If the testing results are so much sensitive to the lighting conditions, then in the real world shooting where the lighting condition can rarely be as ideal as the testing condition, would minor inconsistency in the AFMA results matter at all? I've done a few rounds of tests on all my lenses and I had hard time keeping my lighting conditions consistent so I got slightly different results every time. It was kind of frustrating. So I ended up picking an average value and moving on.
Neuro, et al I'd appreciate you shedding some light on this. Would +/- 5 units be siginificant enough to make any difference in the real world shooting?
Yes, you'd notice a 5-unit difference, maybe not too much with an f/5.6 lens, but definitely with an f/2.8 lens. There are always going to be inconsistencies. Any one shot may be a little off. A properly calibrated system ensures that's cross many shots, the average is at the correct focus. Having the test setup for AFMA as close to 'ideal' as possible (aligned, well-lit, high-contrast target) ensures your AFMA doesn't introduce a systematic error.