Yes, I suspect its the implementation of the canvas object thats the bottleneck here.

An implementation of sub-objects (eg. sprites) on the canvas along with a dirty area only rendering algorithm would go a long way to bridging the gap.