Hi Peter,
Thank you for your help with this.
I'm not resampling functional images to an anatomical template with 1^3 mm voxels. I'm using a lightly customized afni_restproc.py with a custom template for the @auto_talairach step. The template originally had 256^3, 1^3 mm voxels, with a lot of zero-padding. Perhaps only half of the center voxels of any given dimension have brain data. It looks like 3dAutobox has done a nice job, and with the padding option I've left a small buffer around the brain. The space, dimensions, and view have been retained by the 3dAutobox operation.
If any of this seems unprincipled let me know,
Matthew