On the need for off-diagonal terms in the hypothesis of generalized vector-meson dominance