Ouch, this is why I've always done text dumps and have reloaded via psql. At the very worst you can ignore dupe errors of this sort in psql or hand-edit the text dump. We don't use PG's built-in binary large objects but rather store binary data after first uuencoding it, which makes it possible to use text dumps.
This problem isn't really OpenACS related - postgres flat-out is unable to guarantee it will make a dump that will load without tweaking of this sort. It will dump functions that call other functions before they're restored. An obvious workaround is to patch pg_restore so it doesn't die on a dupe function restore.
I suggest you take this to the PG mailing lists to see if Tom or anyone else has any bright ideas.
