Hi, Dan-
You have mentioned "small head motion, <3mm". Actually, I would say that large headmotion can still fit under that categorization.
Are you running the new afni_proc.py APQC-HTML stuff? If so, do you have the enorm+outlier plot that you could show here (or send offline)?
And it looks like you are not censoring points with motion/outliers... that might further lead to suspicion of motion. Consider if you are showing a funny movie, and subjects giggle at the same point-- that would lead to stimulus correlated motion, and something that would show up precisely as high correlation in a naturalistic design.
I will also PM you instructions for uploading the data if you would like me to take a look in vivo.
--pt