Simulation of protein-folding pathways: lost in (conformational) space?