Numerical methods for the QCD overlap operator: III. Nested iterations