These sequences used to be created with "cache 100" in both Oracle and PG but that seems to be gone now in CVS (I think I removed it, I've never seen any evidence that sequence hits are a big problem in real OpenACS installations and the holes are annoying).
I'd check to make sure that the sequence isn't being created that way. As to how that could cause 600-number gaps, each PG back-end would cache 100 sequences so several backends could easily cause gaps of this sort.
