@Magistross: that's what Ash asked in the spec. But here is what I think:
if you use an off-DOM canvas (what you call buffering which is rather caching I think), you will have to use an extra bit of memory for that, and then you still have to draw the resulting big image.
So the question is: what is faster? drawing many little small images, or a big one?
I don't have the answer to that, that's probably a good test to run on jsperf
(maybe if I have time and motivation when I'm back home...)
Also how much place the big off-DOM canvas will take in memory can be a big problem.
And all that is really hypothetically intersting for canvas2D. In webGL, drawing textured polygons should be pretty super fast and way more interesting than rendering a big fat texture that would sit in your memory.
So yeah, the simple answer is that it's a trade off between memory and cpu use. Here Ash prefered lowering memory use.