[eluser]OverZealous[/eluser]
Well, that was easy. Assuming I didn't make any mistakes, I simply changed the replacement regex for subquery to only replace the tablename if it wasn't prefixed with _.
Simply enough, and covers the common cases.
Anyway, thanks (very much) to Cro_Crx for providing such a clear test case and making it so easy to track down!
An updated release with this bugfix, the new language files, and other recent bugfixes is being prepped right now. It should be out in 20 minutes or so.