Peter Holm has tested this patch with a test designed to have many first alloctation requests fail. The result seems to be that the number of successful blist_alloc calls falls a little, and the number of unsuccessful blist_alloc calls falls dramatically. I don't know the details of his test, but since this change should reduce fragmentation, it could just be that having less of the free blocks appearing in 1- and 2- block pieces has a beneficial effect for his test.