Theoretically, now you could render the DOM to a texture instead of to the screen and do some sweet deferred rending.
== Edit ==
Looks like one of Mozilla's Distinguished Engineers blogged about the technique. [1] [2] The creator of emscripten used this technique to render tweets as a texture in the bannabread demo. [3]
[0] https://medium.com/@deathcap1/six-months-of-voxel-js-494be64...
[1] http://robert.ocallahan.org/2011/11/drawing-dom-content-to-c...
[2] https://developer.mozilla.org/en-US/docs/Web/HTML/Canvas/Dra...
The nice part about CSS 3D transforms is that interactivity is maintained, and it's still hardware-accelerated. A fantastic example of this was made by the creator of Three.js over a year ago:
http://mrdoob.com/lab/javascript/threejs/css3d/
As you said above, there are severe drawbacks with that approach. It is very limited in terms of what's possible compared to a traditional 3D scene. Moreover, since CSS 3D transforms are essentially DOM content rendered to a texture, rasterization occurs, and as such there isn't clean scaling, even with SVG content.
var texture = new THREE.Texture(theCanvas);
texture.needsUpdate = true; //refreshes the texture from the source canvas
Now combine it with React.js to avoid redrawing unless necessary and forward the mouse events back to Webkit.js, and you have an interactive HTML texture :)
Impressive. I can't say I see the practical implementations of this, but I'm sure there's some. But it's impressive nonetheless.