The question came up if this was new post-2.4.16. I recompiled 2.4.16 and ran test033 for 30 hours with no failures. I've now switched back to HEAD, and I found first failure in a bit under three hours. (Note that when I found this, I was running all tests, not only 033. With 2.4.16 and today's HEAD run, I'm looping test033 only, because at this point it's the only place I'm seeing failures.)