The computational cost added by the mechanism that decides which experts should process each input in a mixture-of-experts model.