mixed-pitch-mode was toggled *before* text-scale-mode was. They both preform face remapping, but use the same strategy to accomplish it, so the less complex mode should go first. In this case, that's text-scale.