You may be waiting for transaction rollback, but if it is fairly consistent then it is more likely that you are waiting for SMON to do its temporary segment cleanup job. If so, you should see a message in the alert log to this effect. Try using much larger extents in your temporary tablespace. When you get to 8i, use a tempfile tablespace.

Could you please advise what are the checks to be done on the Ora 7.3.4.4 DB to trace the cause of slower database shutdown?