>   syncing disks... 5 done
>   /usr: umount pending error: blocks 21800 files 1124
>   /var: umount pending error: blocks 136 files 4

 I've seen the same thing for /, my only filesystem. I'm running a -current
kernel from this morning, and I have softdeps enabled.  I don't recall seeing
this on my Sparc5, only my Intel laptop. I think I first saw this a few weeks
ago, but I'm not certain.

> I only seem to get this after doing some rather heavy disk I/O (like doing
> 'cd /usr/src ; cvs update -dP').  If I don't do a lot of disk I/O, I don't
> have this problem.

 For me the files that show up in /lost+found seem to be source/headers/man
pages from recent package 'make update's. The dates varied, and some were owned
by odd uids, suggesting they had come out of a tar ball.

> Anyone have any idea what's going on?

 Just a minute ago I reproduced the error by:
  cd /lost+found; rm * and rm one empty directory
  boot -s
  fsck -f / (no errors found)
  finished booting, ssh'd in and su'd to root
  cd /usr/pkgsrc/www/curl (my /usr/pkgsrc was last updated around 8am Eastern
on the 15th).
  make clean cleandir
  make (no dependencies had to be remade)
  make clean cleandir
  'reboot' gave the 'umount pending error' line
  boot -s
  fsck / found a single lost directory and a lot of files. fsck gave me the
directory's parent's inode, but a find on / couldn't turn it up.

 I tried a half dozen more times, with just 'make' or just 'make clean' and it
never failed again, in the 'reboot' or the 'fsck -f'. But a 'make clean; make;
make clean; halt' did fail. I disabled softdeps and that command sequence
didn't fail. I re-enabled softdeps and put a 'sync;sleep 20;' before the halt
and that also fixed the problem.

 That's not a large sample pool, but I think it's softdeps and the large amount
of IO followed by an immediate shutdown that does it.