I have been studying your part 2 data. The final graphs whit the flashes turned ON are of most interest to me. It is very interesting how the flash fires a second monitor preflash if it doesn't receive a return from its first attempt.
I now fully agree with your conclusions that the CLS is much smarter than I realized. There is obviously a composite power being determined after analyzing all the preflash returns from all the groups. I wonder what algorythm it uses to determine the composite power since it has no idea where the flashes are pointed. All it knows is the amount of reflected preflash energy from the camera FOV from each group.
Incidentally, I have noticed that with my D3 it does not underexpose when using remotes like my D200. The D3 is much more likely to be correct. I assume this is due to an improvement in the algorythms used.
Last night I had a photo booth at a Military Ball, and I was using two remote flashes in umbrellas. Most of my images were right on with 0 FEC on the commander. Some needed +0.3 ev and others needed -0.3 ev, but that variation appeared to be due to variation in clothing color.