Or me ...
Yes, I've thought about it, too but we'll have to dump 7.2 support first, right? (I think it's a 7.3 feature)
We could actually have the db_* interface do it as a param'd option, using the fullquery name as the name of the prepared statement. I haven't tracked the details of the prepared statement stuff - do all PG backends see the prepared statement or just the one that executes it? If all then it would be easy to stuff a "prepared statement exists" flag in the XML kept for the query and check it, do the prepare if necessary and then execute regardless.
Worth exploring ...
