This is indeed faster, but it should be noted that the meaning of the code-ref that's passed in is quite different than in the OP... Rather than being a sort function, this version uses the code-ref as a transform used to convert the source values into a bytewise-sortable format. This can be a big performance win, as has been discussed elsewhere, but does require a bit of fiddling to be convert the calling code from the approach used by the OP.