@Ed_S I pushed a patch last week (Oct 29th) that should fix this problem. The problem was that every time you do an upgrade via the web UI, a unicorn worker forked itself 3 times and these forked processes never died. This bug and the one reported here are connected and should be fixed now.
Your instance should have the fix now, so you shouldn’t see any extra unicorn workers after you upgrade via the admin UI. To get rid of the existing dangling workers, you can kill -9 them or a container restart should be enough. Please let me know if you still see this problem in future upgrades.