How two neutrino superbeam experiments do better than one