What you are seeing is an oversight in the way the output parameters are labeled in the special case of '-EPI'.
Normally, the shear matrix is loaded with 3 parameters a,b,c:
[x'] [ 1 0 0 ] [x]
[y']=[ a 1 0 ] [y]
[z'] [ b c 1 ] [z]
and they get labeled as a='y/x-shear', b='z/x-shear', and c='z/y-shear'. However, in the case of '-EPI', the shear matrix is
[ 1 0 0 ]
[ a 1 b ]
[ 0 0 1 ]
so only the output y' coordinate is altered. But the labels for the 'a' and 'b' parameters are not changed to reflect this change in the shear matrix. Hence your confusion. The code is doing what I conceive to be the correct thing, but the reporting labels are inappropriate. I will try to fix this glitch, sometime soon.
You are the first person to ever notice this, or indeed to ever ask about shear distortions in 3dAllineate.