Hi Sam,
1. Yes, even including -polort 0, which is basically required, the mean of each voxel's time series is subtracted out. That will wipe out the contrast.
2. The computations are voxelwise, one voxel at a time. And speed is not an issue. So there would have to be a good reason for using a mask that is more restrictive than the whole-brain mask.
3. Getting into blurs and scaling, it is hard to answer these without knowing what you are trying to do. Is this a full single subject level analysis? Are you using afni_proc.py?
- rick