You can try with Metropolis by passing different variables to different Metropolis samplers and checking if the computation per sample goes down as you are expecting.
If not, could you share more details of your model?
1 Like