I'm experimenting with a change capture architecture for ETL processing that is based on ora_rowscn, and have rebuilt the source tables with ROWDEPENDENCIES to isolate the SCN's to only those rows modified (as opposed to block-level tagging). I'm aware of the extra 6 bytes/row of space overhead, but it is not obvious to me what other impact this would have.
My question: What would be the extra work the RDBMS engine would do with rowdependencies enabled for commits and rollbacks? For my source tables with 100 to 500 rows/block I realize I must be writing 100-500x the number of SCN's (for our typical commits), but are there other side effects I'm missing?