>
> There may be some situations where crawling the indexes a row at a
> time will perform better than this by enough to want to retain that
> option.
If an index existed, wouldn't it still be able to use that in the set-based
implementation? Is there something which would make doing the check
set-based ever worse than row based inherently?