I hadn't read both parts of your message before posting my response (despite the order in which they appear here, you typed faster I guess).
Anyway, my comments about ACS 5 vs. ACS 4 reflect my thinking that most ACS 4 packages probably don't make very sophisticated use of acs_objects and probably won't. I think more thorough redesign and dependence on the object model will be seen in ACS 5. This is only an educated guess, however.
If PG 7.2 has better support for inheritance (inheritable indexes, etc) then I have absolutely no doubt that I'll want to use the feature in any ACS 5 port (which will include a big datamodel port, once again).
And, if PG currently had better support I'd be strongly in favor for 4x.
I'll be very interested in whatever information you can come up with regarding performance issue. I'm busy at the moment trying to grind out a very large set of enhancements to my personal project at birdnotes.net (including making it a lot less ugly to look at) so I can clear the decks for March 1's porting blowout...
