We had a BSD system and an Ultrix system. For convenience I temporarily NFS mounted the root of the Ultrix system into /tmp/foo on the BSD machine. After I was done I forgot to umount the volume.
The next day, the Ultrix machine behaved really weird. Things seemed to work fine, but a lot of files were missing. It took a bit of troubleshooting before we figured out what had happened.
As it turns out, the nightly /tmp cleanup script had run on the BSD machine, and this script deleted all files which had not been accessed in a week. Of course the script had traversed the NFS mount and cleaned up everything on the root disk of the other machine. This meant that all the files that were commonly used were still there, so things seemed fine until you tried to do anything that involved files that hadn't been touched in the last week.
That's when I learned never to use /tmp for anything important.
Of course it was never meant as a place to store users files temporarily, it was always for programs that needed to write a temporary file.