You are most certainly right, @ricardoV94 !
Pure applied coding: when I (used to) set Metropolis in well-defined test cases (back in the days when there were two PyMC’s existing in parallel, you know), back then, Metropolis would go faster. But maybe I am even wrong about that. And apologies if writing “voodoo” offended, I as a frequent user are truly amazed by the improvements in sampler methods and their implementation in pymc, by you guys! Thanks for that!
But discussing samplers was not the point of the OP. Right? I just found the Metropolis response a bit blunt, and thought we might be able to help more if we ask for more info.