Jeff, just FYI...I created the million row table on my slow laptop and tested the code from your article. With a normal SELECT statement it took appx 19-22 secs per run to get a million rows.

Now out of curiosity, I whipped up an iTVF and ran that against the data with a cross apply and it did NOT perform well. Something on the order of 2 minutes or more. That just goes to show that iTVFs are not always the answer. I also re-tested after adding a non-clustered index to the temp table. That dropped about 3 seconds off the ordinary SELECT and about 10 seconds when using the iTVF.

I tried the iTVF both with a WITH SELECT/INSERT INTO and an garden-variety INSERT INTO SELECT and the WITH statement version was the slowest.

Not sure what use anyone might have with that information, but that's my report! 😉