The nice thing about HMC is that it scales really well with dimension. I’ve found that introducing a bunch more parameters isn’t such a big deal as long as the probability mass doesn’t bunch up near constraint boundaries, which can be hard on the unconstrained geometry.