Well, pg_dump's been considerably re-written in the last few weeks, though I'm not sure it will help you.
If you can reproduce this problem as easily as you say (which I know you can, of course), the pg_hackers mailing list should hear about it.
Visit www.postgresql.org and chase links and you'll find out how to subscribe (it's a majordomo list, of course).
There are people there who do nothing but work on PG, and they'll be interested.
