I discovered a way to trigger the test_strptime failure. In 3.6, if I rename the CacheTests class to ZCacheTests in in test_strptime.py, it seems to stop it running before the StrptimeTests class. Then you can trigger the failure it with:

./python -m test -u all -v test_{urllibnet,imaplib,strptime}

1. test_urllibnet loads the _strptime module and initializes the cache with the normal timezone info
2. test_imaplib temporarily sets TZ=STD-1DST, triggering the glibc bug
3. test_strptime tries parsing "STD" using the original time zone settings and fails

You have to stop CacheTests running before step 3, because it resets the _strptime cache.

Anyway, I am now fairly confident my patch will avoid the test_strptime failure as well, so I think it is worthwhile committing it. But I guess a bug should be reported to the glibc people too.