I believe there can be many solutions that would satisfy the initial conditions you have set. The only target set here is the speed of the revolute joint. The large object can have an initial velocity & rotation while the small one staying stationary, or the other way around. Multibody happens to settle at an initial solution that both are moving and the larger one has a much lower initial velocity.
One thing you can try and I attached a model where I did this: start with zero velocity everywhere and apply a torque at the joint after some time. The COM will indeed remain in place.
