Probably fixed here:
https://pagure.io/copr/copr/pull-request/1252

Once deployed in production, we should remove the leftovers and
for some time keep checking that the problem doesn't re-appear.

Still happening:

[root@copr-fe srpm_storage][PROD]# find /var/lib/copr/data/srpm_storage -type f -mtime -7 | wc -l
54

Well, most of the files are in a temporary directory that is no longer referenced
in any of the build in the database. The remaining referenced subset of files
is referenced from builds where the source build failed, e.g. 2123708,
1881047, 1850228 ...

I think that source rpm buidl failure is the trigger, and that people go and
remove those builds that fail ...

Commit 6fa973f fixes this issue