Silly me, you already linked the article I referred to and in one test (the fastest) had the dateparts stored in the table. This removed any date logic overhead that could taint the performance measurement of the function to be tested.

Very nice!