I think the power level for the group is sent and each unit in the group uses the the same value.
The test level for a group would be at a level less then the full power to allow for a fast recycle. And the camera's sensor would pick up the contributing light for each group and compute the needed value for the exposure combining all the groups. And then how is the overall exposure and each group's power level computed? There has been a response from Nikon UK that there is a different algorithm for the pop-up vs. mounted Speedlight. But Nikon UK may not have the real answer.
But what happens when you have SB-600 and SB-800/900 mix since the power levels available for the SB-600 are different from the SB-800/900. My guess is that the SB-600 ignores the value for the 1/128 power level.