Just thought of another option for you. Not the most elegant, and could be tedious if you have a lot of columns, but will work.
Take a SQLPLUS dump (which will take a while) of the tables before and after, to a flat file, and run a UNIX diff on them, or an md5sum.
You'll want to make sure the SQL extract script is very specific, doesn't use any headers or messages that might vary from release.
Just using spool and order by, you should generate the same files.