I think I'd start by putting in some "instrumentation" into the job - just writing out start & end times for each query to a table or log file - that will show you the slow ones. The advantage is that you can leave it there for all time and it will immediately indicate which query is responsible if things start going slow again.

Tracing the session and using tkprof or trace analyser might be the next step - note that this can have a significant overhead. It has the advantage if being 100% focused on the job you want to optimise.

Never seen statspack for a session - is this new in 9i? http://download-west.oracle.com/docs...spac.htm#22046