On the SQL compliance issue: we can do quite a bit to abstract out the inheritance by layering views in much the same way that ACS 4.x currently does to prevent explicit 5-way joins in some queries. Which means that if we wanted to switch to a trigger-version of inheritance, we should be able to do that relatively easily. The PL/SQL abstractions help with that a lot, too.
I wasn't thinking of cutting out acs_objects. I was thinking that, either through views on pg_inherit or through explicit population of the acs_objects table, we would maintain the same meta-data interface.
At the end of the day, I cannot imagine how the internal inheritance mechanism could be less efficient than explicit joins and explicit triggers. Plus, the porting of the data model will be that much simpler (there are a LOT of triggers!!). That's why I'm leaning in favor of it.