You're right. The wedge actually appears after 3dvolreg rather than after scaling. It was more obvious (to me, at least) after scaling, and that's when I noticed it.
The subject nodded back and forth slightly over the course of the run (this is apparent in the original dataset). When I registered all volumes to the first volume, there was no longer data in parts of slice 0 for some volumes because those parts of the brain had rotated out of the frame.
I'm satisfied that zeroing out the bottom slice(s) in all subjects will take care of this. Thanks for all the help!