I thought I responded to this earlier, but apparently not.
First... check the "zoom" setting on your flashes. The should hopefully be set to "Auto". The head on the 600EX-RT and the 580EX II speedlights have a motorized reflector inside which slides nearer or farther from the flash tube to control the spread of light. If you use long focal length lenses then the reflector slides forward and the beam leaving the flash is more concentrated -- the flash carries farther. If the lens is a wide angle the reflector slides back to allow a much more rapid spread of light.
You can set this to "manual" zoom (we're still talking about the flash head ... not the lenses) and you can manually set the focal length for the flash reflector. When you take a photo with a long lens it has a narrow angle of view -- so a more concentrated spread of light will be adequate to light the frame from corner to corner. When you take a photo with a wide lens then the corners would be dark... so the flash needs to spread wider. The focal length values listed on the lens are based on the position of the reflector in order to adequately cover spread the light wide enough to cover the entire frame with light.
Also, you have to be somewhat discriminating in whether or not a ceiling, wall, or other surface is actually suitable for bouncing the flash.
As light leaves the flash, it spreads out.. The farther away from the flash, the less concentrated the light. This causes a light "fall off" effect called the "inverse square law". Each time the distance from the light source increases by roughly 40% (technically the factor is based on the square root of 2 --- about 1.41) the intensity of that light is halved.
In other words a subject at 7' from the flash gets half as much light intensity as a subject that was 5' away (because 5 x 1.4 = 7). A subject located 10' away gets half as much as the subject at 7' and only 1/4 as much as the subject at 5'. You get the idea.
When you bounce a flash, you have to account for the distance the light has to travel from the flash, up the ceiling (if that's your bounce surface) and then back down to your subjects. This will be much farther than the distance from flash to subject directly if the flash were pointed directly at your subject.
But then there's the question of the surface reflectivity itself. Suppose this is a simple painted white ceiling. A "mirror" is a reflective surface so the rule "angle of incidence = angle of reflection" applies. But a painted surface is a "scattering surface". That means when light hits that surface it does not simply reflect off like glass ... it "scatters" in all directions. This SUBSTANTIALLY reduces the amount of light return -- and this assumes it was actually a "white" ceiling. You might lose 2 stops of light bouncing off that surface. For this reason, high ceilings typically do not make good bounce surfaces.
And then there's the question of whether or not the surface is actually "white". If the bounce surface is not color neutral (and preferably plain white) then it will apply a color cast to the light that bounces. If the ceiling is not white (e.g. gray, black, etc.) then it will eat the light.
What I prefer to do in these situations is point the flash directly at the subject, but I'll soften the light if I can. By "soft" we mean that if you inspect the quality of the edges on the shadows... you want a gentle transition from light to shadow. If you see a well-defined edge transition from light to shadow then that's harsh light. Light from a pin-point source creates strongly-defined edges... light from a broad light source creates gentle transitions. So you broaden the light.
Ideally I use an assistant who holds a "side-light" -- a live human holding a monopod mast. The top of the mast has a flash and that flash might be in a small portable soft-box (I use a Lastolite EzyBox Speedlite in the 24" x 24" size. Ideally it would be larger but that gets less portable.) That becomes my off-camera key-light and then I use an on-camera flash (pointed directly straight ahead) but with the power level dialed back so that the off-camera light is brighter then the on-camera light. This causes the on-camera light to "fill in" the shadows created by the off-camera light -- but the off-camera light is itself in a soft-box so what shadows do exist are (a) weak and (b) have nice gentle transitions from light to shadow.
If I don't have a "side-lighters" to hold my off-camera light, then my next best option is a Rogue Flashbender -- in the large size. The flash is pointed straight up (as if to bounce) but the flash-bender is roughly a 12" x 12" square panel and it has a bendable material inside (like a wire but even after a LOT of bending these things seem to never break) so you can "shape" the bounce panel. I curve it into a scoop shape so that I'm really bouncing all my light forward ... but the light originates from the 12" x 12" panel. It's not as good as a soft box -- but much better than a straight-on flash and since we've already established that bouncing off the ceiling isn't really an option the FlashBender is the next best thing.
And lastly... if you must, point the on-camera flash straight ahead but use Av mode. In Av mode the camera will treat the flash as a fill-light. It will still meter for the shot as though there is no flash -- but will fire the flash to make sure you have a good exposure on your subject.
The max flash-sync speed for a camera varies by model. A 5D II and 5D III both happen to have a 1/200th sec flash sync speed. This means the shutter speed needs to be at 1/200th sec OR SLOWER to expose with flash. But that's probably not a long enough shutter speed to collect much light from the room. This causes the room itself to look much darker than it really way and the subjects look bright -- it's a very bad look. So the camera compensates by slowing down the shutter speed to allow for more of the ambient light exposure. This causes the room to look much more natural. When we slow down the shutter speed in combination with a flash it's called "dragging the shutter".
This means, for example, that your camera might use a 1/60th sec shutter speed even though your flash is actually only lit for a tiny fraction of that time -- the rest of the time the shutter remains open to collect ambient light from the room.
Here's an example:
This is straight-on flash... no soft-box, no bounce, no Rogue Flashbender, nothing... just plain "straight on" flash -- the kind that usually doesn't get good results. But instead of bright subjects with a very dark background, the whole scene looks rather natural. That's because the camera was in Av mode set to f/2.8 at ISO 400 and the camera used a 1/60th sec exposure time (even though it could have used a 1/200th exposure.) This allows for the camera to also expose the background in this room and that's why the lighting looks good.
Incidentally the ceiling in this room is VERY high (at least 20-25' up) and it's all wood paneling (not an acceptable bounce source.) That's why I used straight-on flash.
This shot works because the background is FAR behind the subjects. I've rotated the camera wideways and that means the flash is on the left side... this would normally cast a visible shadow which you would see to the right. But since the walls are so very far behind the subjects you do not see this shadow (if it were closer you would see the shadow.) If the flash were above the camra (horizontal orientation) then you wouldn't see the shadow at all (it would be completely hidden behind the subjects.) There are flash brackets that keep the flash above the camera at all times and the camera is able to rotate to a vertical orientation on the bracket without moving the flash. They are designed for exactly this type of situation (straight on fill-flash but don't want to see that shadow.)
If you look at the external flash control menu on your 5D III (and in your 5D II I think this is in custom settings) you can set how the shutter speed works WHEN the the flash is used specifically in Av mode. The default is "auto" which means the camera will pick any shutter speed it wants as long as it's at or below the max flash sync speed. But I set mine to "1/200th - 1/60th" which means it has to pick a speed specifically within that range (it's never allowed to use a shutter speed slower than 1/60th). You should set the ISO to Auto when you do this so that if the camera must boost ISO to get the exposure correct then it will (it will do that as a last resort... it'll try to keep the ISO value low.)
I appreciate the detailed response. But I am still trying to find an answer as to why a picture shot in low light, using bounce-flash with a lens that has a maximum f2.8 aperture is underexposed by several stops, while a picture shot from the same location with all the same settings but an f1.8 lens SET TO THE SAME f2.8 aperture is perfectly exposed. Here are some examples with the first of each image shot with a 16-35 2.8. The second version of each scene shot with a 50mm 1.4 or a 28mm 1.8. The same underexposure happens with a 70-200mm 2.8. Happens with multiple camera bodies, and happens to other photographers I work with as well. In some cases, the underexposed 16mm photos are shot with the FEC on the flash dialed all the way up to +3 stops, but are still underexposed.
These are all taken at high ISO (3200-5000) and the rooms are very large, ceilings are extremely high and aren't necessarily white. My suspicion is that the pre-flash is having to discharge significantly more of the capicitor to get a reading with an f2.8 lens than an f 1.4 lens, so there is not enough charge left in the capicitor when it comes time to do the actual exposure. This would only make sense if the preflash occurs BEFORE the camera closes the lens down to f 2.8 for the exposure. Does anyone know the sequence of events for pre-flash on E-TTL II? Does pre-flash occur while the lens is still wide open?
