Hmmmmm... Based on what I saw of the article, I'd have to say that the conclusions drawn were unwarranted. I didn't see anywhere where the 3 different queries were proven to be equivalent, nor did I see the data volumes being worked by the queries. I don't think you can draw the conclusion from the evidence given that CTEs will never write to tempdb. Could be true, though counter-intuitive, but I don't think the article proved it.