The problem now is that ls now DOES a stat (lstat) for every
   file in a directory.  This may not be terribly noticeable
   on a local filesystem with a hundred or so files, but it is
   VERY noticeable on an NFS mounted mail spool with about
   7,842 files.  I guess with the zeal to do ls "properly,"
   performance becomes impertinent.

huh?  ls(1) only does a stat when you ask for info that
requires it (such as -s, or -i or -l, etc).  i just tested
it with ktrace and there was only one fstat() call and no
stat() calls.  this has always been the case as far as i'm
aware.

.mrg.