When using the .mod generated from cml (after replacing the appropriate parameter gmax->gbar permeability->pbar and shifts) in the original model, the simulations run much slower. Evidently, the .mod from cml have much more RANGE and ASSIGNED stuf, as well as intermediate vars, but I would expect those to be optimised off after compilation.