I have to correct my prior message. Postgres isn't crashing during vacuums -the vacuums are just going on indefinitely, eating 100% CPU.
I tried manually issuing the vacuum command and it did the same thing...