It seems the oblique transformation matrix stored in the header is not correct for one or both of the epi and anat datasets. The alignment script should work anyway, but you should set deoblique off. 3dSkullStrip may take longer for some datasets and not display any output. You can also run this step separately for the anat and epi datasets outside the align_epi_anat.py script if you like. The visual mode described in the 3dSkullStrip help is somewhat entertaining while you're waiting.
Note you may not be able to see subtle rotations easily in the way you have shown here. There is only about a 2 degree difference here between these two datasets. Instead, try lowering the opacity or, even better, using the -AddEdge option to the align_epi_anat.py script.
So you might try something like this:
align_epi_anat.py -anat s2-3d+orig -epi s2-runall+orig -epi_base 5 \
-deoblique off -epi2anat -suffix _al2anat
3drefit -deoblique s2-runall_al2anat+orig