I am beginning to suspect that the osx box I have been using for testing is actually tossing at least some of it's data into the VI queue, and that is the source of the rrul_be asymmetry. I don't know if I can take actual caps of the 802.11ac air with an ath10k with three antennas at the moment.
As this "optimization" only grants the osx box more bandwidth for itself, and most (but not all) of the
current airtime scheduler is optimized for switching between stations fairly, I think if I add another non-apple box to this test, that I'll see a result closer to what I would expect.