I am not near my workstation to test, but I am curious if the tests were affected by caching? Any time I have two statements that are different but produce the same timing I am suspicious. I know that these statements may actually run the same in the engine, using the same query plan, but then again the difference might be in the details there.

Perhaps there is a difference in how the query performs when there is no data already in cache. I know that this has me interested enough that I want to test this against a massive dataset, then retest with the old query after cycling SQL.

Great article though, this may prove very interesting.