An image is missing from http://ge.tt/88JfXx4 due to what seems a wrongly calculated mask bbox for which an image's ctm instead of the pre-image ctm is used. Shouldn't pdf_begin_group use gstate->softmask_ctm instead of gstate->ctm for calculating the softmask's bbox?
Our fix: http://code.google.com/p/sumatrapdf/source/detail?r=3895
Fixing this bug causes the problematic file in bug 692312 to crash with a segfault in fz_draw_end_mask.