Thank you very much for your links! While looking through your blog I also found: https://junpenglao.xyz/Blogs/posts/2017-11-22-Marginal_likelihood_in_PyMC3.html
This is basically answering my question even more directly on how to get to the model marginal probabilities and then to the bayes factors. Great blog posts!!
By the way: at the bottom of that post you say: “I guess it is time to double check the marginal likelihood estimation in SMC.” Did you find the root cause why the different methods deliver differing results? Just out of couriosity. My core question is answered now.