It seems the scheduler is not always cleaning up old recipe repos properly. On our Beaker instance we found 591 old repos under /var/www/beaker/repos dating from July 2010 to one week ago. They are "old" in the sense that the corresponding recipe is already completed, cancelled, or aborted, so the repo should no longer be present on disk.
A large number of the recipes in question were cancelled -- it is possible we are creating the repo and not destroying it, if the recipe is cancelled before it starts. (See also bug 715226.)
Bulk reassignment of issues as Bill has moved to another team.
I think with all the fixes for task repo generation and job status updating in recent Beaker releases, this issue should be fixed. There are no orphaned repos in our production Beaker instance, aside from a handful from May 2013 (which pre-dates all those fixes).