Duplicate rows. Not many, just enough that a daily report came out wrong and somebody noticed. The data looked like the import had run twice, which made no sense, because the import was a cron job that ran once a night.
It ran once a night on each of two hosts. We'd cloned a box for capacity and faithfully copied the crontab with it, so at 02:00 both machines woke up, both decided it was their job, and both ran it. They didn't conflict in any loud way. They just each did the work, and the second one to finish quietly created the duplicates.
The proper fix is flock, not "remember to remove the crontab from the clone", because I will not remember.
0 2 * * * /usr/bin/flock -n /tmp/nightly-import.lock /opt/jobs/import.sh
-n means non-blocking: if the lock is held, the second invocation gives up immediately rather than queueing. Whichever host grabs the lock first does the work, the other does nothing, and the duplicates stop. Cron has no idea your job is running anywhere else. If it must run once, you have to say so explicitly.