It's also a display model problem. A big bitmap that you draw on arbitrarily is just harder to optimize than a decomposable scenegraph.
This downside of Canvas also means it may become a dead-end for high-performance 2D, unless we discover some way of making all manipulations go completely on-GPU. Canvas is presently best suited for creation of individual textures and sprites, while other methods(CSS, SVG, WebGL) are better to render and transform them.