On the validity of the ray approximation for interpreting delay times