I haven't read the article properly and just scanned it to see what it is about and if it is interesting.

But one thing I noticed quickly was a quote at end saying:

Which fired all alarm bells in my head at once as one shoud not use set statistics time on for any serious benchmarking, ever! Measuring performance this way can have a really big impact on the actual execution times themselfs. This has to do with how SQL Server internally processes various constructs and its points of measurement.

The most reliable way of measuring is to have long sufficiently long runs and start with a GetDate() and end with a GetDate() and compute the difference after the test has run. You will require a variable to store the start time for this, but it won't affect the results as much.

Other then that, the article looks as quite a bit of decdication and work has gone in, and I will sure read it in full when I have the time. In the meantime, please recheck the conclusions based on this different method of measuring, you might be surprised!