On the optimality of linear methods for operator approximation in convex classes of functions