> There is some kind of serious bug here.  Things should run slower if you
> fill up the vnode table, but they should still run -- the table should be
> managed by LRU.

Hmm...  I have a setup with a pile of null and union mounts, and I wound up
doubling kern.maxvnodes (some months ago, in 1.6 at first) so I would stop
getting failures.  I assumed that was Just The Way It Is and didn't keep
details of the failures.

-Chap