After experimenting with some of the more familiar approaches to permission_p scaling (such as restricting the number of rows that it is run on and avoiding placing it in a WHERE clause), we've started to look at its internals. A proposed change is mentioned in this thread: http://openacs.org/bboard/q-and-a-fetch-msg.tcl?msg_id=0004rw&topic_id=12&topic=OpenACS%204%2e0%20Design. The thread did well for a short while and then sort of petered out (no pun intended). We have yet to carefully test the proposed changes but you might want to take a look at them.