Not a scientific / empirical answer (i.e. I haven't done side by side comparisons), but I would think that in-camera noise reduction would give better quality.
As I understand it, the in-camera noise reduction works by taking a second exposure and reading the noise off the sensor, then taking out that noise as best as it can from your first exposure.
Because that second exposure is in/with the exact same conditions (exposure time, ISO, temperature, etc, all of which affect sensor noise), its more accurate than a post-processing algorithm that has to try to interpret the image and decide what is noise, and then remove it.
