Problem solved, we think. We had a couple of reverse proxy servers set up that were hitting the wiki constantly with keepalives. Once we turned those off, mono CPU dropped down to zero (or close to it).
I'm still not sure I understand why the high level of activity would have caused the wiki to keep crashing periodically, but we'll see if that stops now.
Thanks for the help, sorry for the wild goose chase. I meant to check the Apache logs earlier, but somehow never actually did it.