Here is some new data. I had the FTDI cable connected, and seems like it was influencing things somehow. I cut the RX (always high ~ 3.5V) and that changed the way Q1 is turned on. Still not sure why. I removed the FTDI altogether and repeated the readings.
The Mayfly will now continue operating to V_batt = 3.00V, Proc resets when I drop to 2.95V (motioning with a reasonably professional Meterman 37XR voltmeter).
The overhead/buffer difference V_batt &rail3.3V is now about 0.15V, probably mostly the SPX3819.
However it does show that the mega1284 A6 measurement is only linear to about 3.5V.
The 2nd graph shows the difference between V_Batt and 3.3V rail
I’ve added the .xls with the readings.