The relax application basically performs a series of repacking and minimization with ramped values of Lennard-Jones repulsive weight. As far as I can tell, the only Monte Carlo-like portion of the protocol is the repacking step, and that’s more simulated annealing than straight Monte Carlo. For the most part the temperature and ramping of the packer’s simulated annealing is all handled automatically (it’s not something that even most Rosetta developers bother messing about with).
If you’re interested in getting more aggressive conformational changes, my (admittedly naive) impression would that changing the repulsive ramping scheme in relax would likely be more beneficial than messing around with the packer’s simulated annealing. The relax documentation has information on how to specify and adjust a custom relax script. The defaults have been optimized quite a bit, though, so I’m not sure how much playing around with the script will get you.
.
