Just to be sure, I appreciate anyone willing to step up and write articles to share their knowledge with other people especially when they come up with such a different method.
Getting to the point, though... I expect no performance gains just because something is an SQLCLR type behind the scenes. In fact, because of things like the FORMAT function, I've grown to being extremely skeptical when it comes to expectations of many built-in SQLCLR backed functions.
With that being said, I was looking for some form of performance testing (or the results of such testing) in both this article and the one you cited above. I'm just not seeing anything that might resemble a test harness. To be sure, I normally test things with a million rows. In this case, I would probably test with a million intervals that contain at least 50,000 gaps of 1 to 1000. Then I'd do the same thing with some of the more commonly known methods.
Who knows? You might be right. Maybe the SQLCLR code behind the scenes was written much better than things like the FORMAT, et al, function.