Well, we don't want to use the current LOBs in Postgres because, quite frankly, they are horrible. We're trying to use varchar for text LOBs since with compression and larger blocks, we can easily get 32K of data into a varchar, and that's really all we need (a textarea in Netscape is limited to 32K). As for BLOBs, we're using Don Baccus's latest hack which "simulates" BLOBs. This is important because current LOBs don't get dumped in pg_dump operations, and Don's hack uses normal byte array fields which is much more interesting from a backup standpoint.
Now, your interface will most probably prove very useful once Postgres moves to interesting LOB functionality. I'll let Don comment on the details since he is "the man" for this kind of stuff.
