Hi Terry,
Thanks for the article.
It's definitely an interesting approach - but I miss an explanation. WHY does the second version run faster than the first? I would have expected the reverse, but your measurements disagree - there really should have been a paragraph or two to explain this in your article.
I tried to verify your findings on my computer, and I found that each version ran in subsecond duration; your tests must have been done on much older hardware, or on much larger tables. I'm a bit short for time now, but I plan (if I find the time) to do some more experiments, maybe on larger tables. For now, I can only say that the execution plans do not support your observation that the second query is faster, and that the execution times are too short to be sure.