The results should not vary except at edge voxels that are either included or excluded in the autobox mask. 3dAutobox uses a median image across sub-bricks to compute an automask that it then uses to determine the size of the box to limit the dataset. It should not change any voxel values other than eliminating or including voxels entirely on the edge of the volume. That may affect the scaling of the overlay image and, consequently, the color assigned to each intensity. You can check the voxel values themselves at any x,y,z location and not just the color with the Jump to xyz function in the AFNI GUI and see the value in the Overlay panel. Alternatively, you can fix the scale for both overlay to be the same by turning off the "autoRange" and setting the same maximum for both datasets manually.

If you would like to use the same mask for each dataset, you can apply the autobox from one dataset to another with 3dZeropad -master, or create your own automask with 3dAutomask. 3dAutomask has options for dilating the mask. 3dAutobox can then be used with that dataset instead to create a slightly larger mask. You can then reduce the dataset size with 3dAutobox -noclust, and then 3dZeropad -master.

3dAutomask -dilate 2 -prefix dset_am dset+orig
3dAutobox -noclust -prefix dset_am_ab dset_am+orig
3dZeropad -master dset_am_ab+orig -prefix dset_ab dset+orig