It appears that the spatial information in the IBSR 2.0 is inconsistent within subjects.  IBSR_01_segTRI_ana.img, for instance, has an translation in its transformation matrix that doesn't appear in IBSR_01_ana.img or IBSR_01_seg_ana.img.  Some subjects also have a difference in voxel size.  The voxel-by-voxel correspondence in the .img files appears okay, so a workaround is to use each subject's anatomical .hdr and .mat for the segmentation and tri-segmentation.  (Andy, can you verify this workaround is sound?)