It's not that complicated:
You have two synths inside... each having either 4 or 8 voices depending on what version you have of the REV2.
Layer A is basically synth 1... and it is hardwired to output A.
Layer B is basically synth 2... and it is hardwired to output B.
BUT... if nothing is plugged into output B, then output B is mixed internally in hardware with the output A (this cannot be changed in software).
That's it... there is not more to it than that.
What seems to confuse people are the fact that "Layer A" playing mode (16 voice mode) is labelled exactly like that ("Layer A")... that makes people think that playing mode "Layer A" will ALWAYS come out of output A...
but that's just not the case... in 16 voice mode it alternates between playing a voice on synth 1, then synth 2, then synth 1 etc... that is why you hear it bounce from output A to B like that... it needs to do this, otherwise you cannot get to play with 16 voices in the "Layer A" playing mode.
What you need to understand is, that the playing mode labeled "Layer A" is in no way referring to a specific output.. .it's referring to the preset's data for layer A...
What really happens "behind the scenes" is that whatever sound is programmed in Layer A when you use 16 voice mode, is actually copied to both synth1 and 2 internally, so that both synths inside have the same sound configured in them... you just do not see this, it's happening behind the scenes.... it's basically two synths playing the same preset (in this case both are playing the data from layer A), each taking a turn at playing a voice alternately to achieve 16 voices of playing.... thus, if you have anything connected to output B, anything that synth 2 is playing will come out of output B.
Hope that clarifies it