Really nice to see this topic discussed!
Being mainly experienced in CAE for mechanical, control, and some hydraulics (OpenModelica, Adams, Simulink, "raw python"), I've read this with great interest.
I don't want to limit the discussion, but my take is that if such a thing has a chance it would need to "pick the low hanging fruits" first.
Since FreeCAD is first and foremost the tool to take care of geometric definition of systems (i.e. mechanical and dimensional properties), a primary focus on that would probably yield the fastest results. In that respect I feel that @looo seems on to something with the topic "node editor".
When modelling a 3D mechanical system in OpenModelica (i.e. chiefly using the Mechanics libraries Multibody, Translational, Rotational) the main benefit of FreeCAD would be to define coordinates, body types, joint types. In fact, especially bodies with non-trivial geometries are almost impossible to model in OpenModelica directly, so FreeCAD would be a big win there, even if it were only in a simplified manner with lumped masses and inertias for a start. Then of course animation of the results is what we are doing it for, but this should be "the least difficult", since the difficult topic of kinematics would be taken care of entirely by OpenModelica (the simulation result) in this case, so you would "just" need to force result coordinates and attitudes on the single CAD parts for every time step during animation.
Another topic (which was mentioned here as well) could be for geometrical properties of hydraulic piping, but there I suppose it will be more of a pre-processor use case.
Then again, for multi-body problems the approach @josegegas has taken with MBDyn seems to have taken a great head start, and come to think of it the things MBDyn needs as an input are quite comparable to the things OpenModelica would need, so synergies would be promising there I suspect. I have been wanting to find the time to take his toolbox for a spin...