Clearly, the low yield tails decrease as the FR size increases. Some of the low yield samples at 3x3 require the presence of rare bubbles as large as 5mm. Whether that shows up as noise depends on whether the bubble flies thru the field of view in 1 msec (very noisy) or grows and hangs onto the window for many samples (not very noisy). Low yield tails don't necessarily imply a noisy destiny.
I wanted to blame the 0.5% increase in yield at 5x5 on rising beamline backgrounds. That would represent a not-unexpected doubling since 5x5 is the limit of the "safe box" beyond which the corners of the multiple-scattered primary beam start to clip the W plug. But that hypothesis won't accomodate a similar sized DECREASE in yield at 4x4. Since this was a quick test, perhaps the time for the PMTs and target to stabilize wasn't always controlled.
Anyway, really interesting! A plot like this, with more statistics, has GOT to go in someone's thesis and be correlated with potential increases in whole detector statistical noise (be it Lumi or MD).