A remark on natural constraints in variational methods and an application to superlinear Schrodinger systems