> /home is a mount point.  [...]  But it should be noted that the
> postfix, postgresql, and apache users don't touch /home.  postgresql
> resides under /var (which is directly under /), and they exhibit the
> same kind of error.

The closest common ancestor is, then, /, right?

UTSLing makes me think that fstatvfs1() - the syscall that fails in
your ktraces - does stuff walking up/down the directory tree, unless
certain flags are given.  I'm beginning to suspect that something is
weird with / when the system is hosed.  It sure looks to me as though
the errors occur for, and only for, things that read / (as opposed to
searching it), as if it loses its world read bit or some such.

The real baffler to me is how logging out fixes it.  Do you have a
program that just does a stat() and prints out the results?  Siccing
that on / might reveal something interesting.

/~\ The ASCII				der Mouse
\ / Ribbon Campaign
 X  Against HTML	       mouse@rodents.montreal.qc.ca
/ \ Email!	     7D C8 61 52 5D E7 2D 39  4E F1 31 3E E8 B3 27 4B