oh, in that contest, being able to parallelize allocation this way must be a big win. Of course it wouldn't work with a free list.