Canvas and WebGL are fundamentally not the same thing. The Canvas API specifies a bunch of behaviors (some of them very expensive to implement) and it also specifies an API that is inherently inefficient and poorly suited to what most 2D games do.
If you were to benchmark a WebGL implementation of the Canvas API (to spec) against Canvas, that would be realistic. You can do that, there are a couple out there - but you might find that they don't really perform much better. WebGL2D certainly doesn't; it's slower than Canvas in many cases.
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.
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.
Also, the Web Audio API is available to all web pages in all recent stable Chrome builds, as well as Safari on iOS 6+ - it's not just a Chrome experiment. The Web Audio API, along with WebGL, are both very good things for HTML5 games, and especially larger engines like ours. I think they're both ready for prime time and the fact other browser makers haven't got round to it isn't any excuse not to support it. Why can't IE be a browser that leads the way rather than leaving Chrome and Firefox to invent everything?
Why is it stupid to put canvas 2D head-to-head with WebGL? They're both rendering the same thing, and it illustrates how WebGL is far faster than canvas 2D for drawing the same content. That's an important consideration for a HTML5 game engine, no? Under the hood most canvas 2D implementations plug in to a 3D API at the same level as WebGL, and it shows how cutting out the middle man can win you lots more FPS.