Ah... I discussed this approach with davido the other day, I thought the benchmarking showed that this was actually slower than my simple approach. Strangely enough. ;-)