No, I think he's simply saying some subjects have more aptitude for the topic than others, and this distribution may be bimodal - explaining why some results are good and some are bad. None of these studies address whether the epistemic debiasing leads to improved or worsened instrumental results. (That particular idea, that improved epistemic rationality can lead to decreased instrumental rationality, would be the 'Valley of Bad Rationality'.)