Should N-body integrators be symplectic everywhere in phase space?