On γ 5 in higher-order QCD calculations and the NNLO evolution of the polarized valence distribution