The situation is also way above me - I have never seen Oracle behaves this way... For me this was always the most scalable RDBMS!
I have traces, of course. Bellow is the summary of the last one done for a single app instance. No triggers, no FKs, a couple of indexes on the big insert tables as every session also selects from them.
Both selects and inserts are intensive. It'a heavy OLTP (huge insert rate, small multiple selects, etc.) but imagine that the Db is treated as a huge repository - for all records inserted and evaluated. Every new records is compared towards the old entries. If existsthen the re-evaluation is performed, If not - it is inserted as new.
Thanks a lot,mj