I agree, all of HTML5 is in beta, but lots of people are working on game engines. I'm doing one myself (which I'll release open-source and describe fully in a book on HTML5 game dev.)
So far, I've got a basic sprite object that works much like a MovieClip in ActionScript or a Sprite in PyGame. It's got all the basic features: speed, direction, vector-projection, multiple images, boundary-checking, transformations, and basic collision detection.
I'm working on a very simple game object as well. This will encapsulate the canvas element, manage the main animation loop, and provide high-level interface to the event system (such as it is)
I'm trying to keep it as simple as possible: something like gameEngine in Python:
Even when I get this working, it certainly won't be stable, because the underlying technologies are far too unstable. The browser support of various HTML5 features literally changes every day, and there is still no meaningful support for most HTML5 features in IE. Also, performance varies dramatically between browsers, so a game that runs great in chrome may not run at all in Firefox.
If you're wanting a stable, practical environment, HTML5 isn't it yet. If you're interested in experimenting with HTML5, it's looking like a very promising alternative. I'm definitely eager to use it as a teaching environment. I believe it will be pretty useful in that regard.