On a separate yet related subject, There are various techniques that may improve the model simulation times. Here are some of my tips.
1. initial conditions - watch especially starting a joint at its end of travel, in or near the "coefficient of restitution" region. Sometimes moving the starting configuration a tiny amount away from these end conditions can dramatically reduce simulation times.
2. divergence - this often happens with dynamic models. It takes experience to avoid/correct this. Making a model more realistic but without adding too much complexity helps. Add a realistic small amount of damping like the real world. Really high friction also is very unstable. (start low like mu 0.1 to see how that affects simulation time)
3. accuracy, sometimes increasing the accuracy actually simulates faster due to error reduction. (File - prepare - model properties - mechanism -relative tolerance, and characteristic length) I experiment with various tolerances from 0.001 to 1E-6 to see how sensitive results and simulation times are to the tolerance.
4. time step: dynamic events may need very small time steps which adds to a long simulation time. Adding mass/inertia is another way the model can be dynamically relaxed to allow longer timesteps and longer simulation times. Use a shorter simulation time if that represents actual use cases rather than slow times like laboratory measurements are often unrealistically done.
5. 3D contacts. Replace with CAMs wherever possible These can be used for contact in a lot more cases than you might think. Normally you think of the cam and follower happening in a single plane, however I have had good success with a perpendicular relationship (similar configuration to two 45degree bevel gears having perpendicular axis they rotate around) My experience is that CAMs simulate much faster than 3d contacts.