Some remarks on two-loop superstring calculations