ianwood, here are a few more things to give some consideration to.
The GPS module uses battery backes RAM (BBRAM) to store last fix info, TCXO Calibrations, Almanac Data and RTC info.
The module requires a voltage above 1.8 VDC to retain this info
If the backup battery is below that threshold, every flight is a cold start. The TCXO calibrations and Almanac data download are time consuming. And the NAZA may not have the computing power to deviate from its task manager cycle to assimilate the most up to date info immediately. It may have to cycle back to he read timer tic sequentially, meaning latency in getting correct compass and GPS info.
Now here is the kicker... I have never measured a backup battery in my Phantoms above .5VDC.
The compass data goes straight to the GPS module, from the lowest point to the highest point through a 5 wire ribbon cable near 9" long. That is nearly 4' of antenna to capture any interference and further complicate. I don't think that is in pay in the numbers you are gathering, but it could cause enough of a deviation to mask the true numbers, if interference is factored into the mix.
I believe you are seeing the TCXO and real time clock ((RTC) calibrations in real time, due to low backup battery voltages. The wider the declination, the more complex the calibrations.
The references to lookup tables makes me think that others have had to second guess OBD II routines in the past also.