I show that, in some cases, INSTEAD OF triggers can be better for performance than AFTER triggers.