Indeed; the real speed gain comes from the iterator-like workings of each -- no need to build a list of keys to feed to foreach before even starting the loop (and at 20k keys, that costs quite a bit).