The MNI space is larger and shifted away from Talairach space. The MNI 2009c template is also different than the N27 dataset in MNI space, so the regions that correspond to MNI 2009c template space are not going to have direct x,y,z correspondence to TT_N27 space. If you look at the same region on the MNI 2009c space, it's much more sensible. Still we have found that region specifically has a known problem and needs some correction by the authors. The posterior end of the region goes too far superior. We are waiting for that correction. In the meantime, you might consider the similar regions in the MNI HCP Glasser atlas and the Brainnetome atlas. Both of those atlases show regions similar to BA22 as two separate regions.