You may have already decided but I would second the pinning the animation object to the physics object.
Physics can be touchy. If you are animating a physics object you are constantly changing the size and location of the object. This could cause unexpected results from the physics calculations because:
If part of the object is animated from one location to the other it translates into very high speed and could react like a high speed collision
If the animation changes sizes (of the collision polygon) it affects the mass of the object (size x density). So if one animation had arms and legs sticking out (running or something) and had a bigger collision border it would have more mass to throw around in that instance.
Not to mention an animation causing collision polygons to suddenly overlap.
I would pin the animations to an object with an unchanging polygon or else crazy things could happen.