Yes I agree and initially these were my thoughts exactly. If I a start with all cells at the same exact voltage, then draw same exact power from each cell, then recharge them all with the same exact amount of power each, they would arrive exactly back where they started from- all at same voltage.
However, there must be something more in play here with the physics of the energy flow which I am not comprehending. It could be the assumption that the same amount of energy is leaving each cell, which may not be true. For instance, if cell #3 is at a slightly higher capacity at the beginning, it will always be at a slightly higher voltage than the rest of the cells, which means even if amps are the same, different watts are leaving the cell. There may be other laws of energy working here that I don't understand either.
Another idea is that during top balance I am at very tiny amounts of current, however finishing a charge cycle with bms hooked up I am charging at approximately 30 amps, which could thus exacerbate differences between the cells which would not be apparent during the trickle charge top balance. Also, I am discharging at around 93 amps and charging at 30 amps, discharge could be a different efficiency out of the cells vs charge efficiency into the cells which may affect cells of varying capacity differently. There may also be a slight peukert effect affecting cells of differing capacities differently.
I finished a charge cycle last night at about 25amps up to high voltage cutoff and all the cells were within 15 millivolts which is pretty close. Today I will do the same and do a capacity test for 3 hours down to low voltage cutoff and see if i get any more capacity out of it.
Meanwhile, on my other assembled pack, I have so far completed individual cell capacity testing (at 20amp discharge rates- the max of the tester i'm using) and so far have gotten 285ah & 282ah from the first two cells. This is all pointing to two things- 1) Just as Docan Power stated, the higher the amp draw you test these cells at the lower their capacity (characteristic of cells that failed to make EVE's Electric Vehicle grade) and 2) an assembled pack has much more parasitic loss through bms and associated wiring which is not measured at my shunt.