Thanks for the explanation - its a fun program, and despite all the warnings, isn't even as slow as I expected it to be. So far no issues with this, and it seems to be behaving as I expect it to.
Am I correct in understanding that this time series correlation approach is fundamentally different from the approach used by 3dLocalStat -stat fwhm -nbhd XX? I'm clear enough on the reason for the ACF and its corresponding long tail - but it seems that 3dLocalStat FWHM doesn't care so much about time.