Improved renormalization of lattice operators: A critical reappraisal