The collision detection could be improved by writing your own set of solidity classes, of course, which would be much more efficient, I think. It doesn't need to be perfect, obviously, since the two sprites cannot overlap their bounding boxes by any significant amount and still not be colliding. You could just have a class that makes the tile only half-solid, or three quarters, or even a trapezoidal shape so you can be somewhat forgiving. I don't imagine you need your sprites to be so very flexible such that it cannot be represented by animated tiles. In the odd cases, you can still use your sprites.
