Performance measurement is hard and expensive to do right. I'm unconvinced by this. A few points.
1) I want realistic tests of what I'm doing. These tests don't measure up.
2) .NET has ticks with 100 nanosecond resolution. Is that not good enough?
3) There are other ways to use XSLT in .NET. Is this a realistic choice.
4) How does .NET 1.0 compare with 1.1?
5) I imagine the choice is more like ASP versus ASPX in which case running msxml from .NET is inappropriate.
6) Where are theunits for the axes on the graph marked. (This is really basic!)
In summary. I'd love to see analysis of optimised code for each situation tested. I admire someone taking the time to have a shot at it but am unconvinced. At present I stick with .NET XSLT within a .NET application.