What single experimental requirement makes superconducting devices expensive to operate? In principle, can this limitation be overcome?