I simply think it's a misleading comparison. You could easily compare Canvas performance in Firefox and Chrome against Canvas performance in IE and the better browsers would probably still come out ahead. If you wish to simply demonstrate that WebGL is faster than Canvas, you should do that comparison in the same browser. You could also demonstrate some of the use cases that simply aren't possible with Canvas, or are much slower (blend operations, for example).
My perspective comes from porting a bunch of XNA games to HTML5, so I've spent a lot of time digging into the semantics of Canvas and trying to figure out how to use it to achieve precise results in varied scenarios. That's part of why I think it's inaccurate to treat it as equivalent to just drawing textured quads with WebGL. A lot goes on under the hood.
For the record - the different rendering with WebGL was a separate point about supporting shaders, the performance measurements were all based around identical-rendering tests.
I think our perspectives differ because my view is different to yours: we develop a large, does-everything HTML5 game engine, and I suppose you're arguing more from the point of an independent developer hand-coding a game. So I can see where you are coming from, but I still disagree about WebGL. Our engine doesn't care about how anything as drawn, as long as it looks right. Both WebGL and Canvas 2D achieve that goal, and one happens to be faster. Since we don't need two identical interfaces and can happily maintain two renderers, this is a win for us. In fact, WebGL can render absolutely everything Canvas 2D can, just with a different (and still standard) API. I think it's fair to compare the performance of two different APIs, especially when they are drawing identical content.