I’m afraid we don’t have any “best practice” tip when it comes to softbodies.
About your first question: yes, it makes sense to have two different models, one high-poly for visualisation, and one low-poly for physics. This makes sense for rigid bodies too: collision detection for triangle soups (arbitrary triangle meshes) is rather fast in modern physics engines. However, it is still a good idea to have two different meshes, one for visualisation, and one for physics. Best example would be the static environment (ground, houses, trees, etc.).
About your second question: I can’t give much advice here, sorry. Our wrappers just expose the native Bullet methods when it comes to these parameters. It might be best to ask for help on the Bullet forums. Describe the effect you want to achieve, and maybe someone can give advice on which softbody configuration to use.
About your final question: Yes, a rigid body in between might be a good idea. Another idea would be to use sb_node.append_linear_joint(…). Linear joints have been added only recently, so you have to use current snapshot builds. See this thread for more information: Bullet appendLinearJoint to attach softbody with rigidbody