I think Dave's right in that this is the only difference. If you're playing around with these groups do a CVS update from the oacs-4-5 branch, I've fixed bugs in their creation in both the Oracle and PG version. The PG version was thoroughly broken (an inverted tree query, i.e. one that ran in the wrong direction from the start node) and the Oracle version didn't work if you had a subsite tree more than two levels deep.
