I'd be more interested in seeing the execution times where the data size is non-trivial. My suspicion would lead me to believe the row_number() solution might scale better based...