The paper purposely handicaps the vegan simulation to only use 70% of arable land so as to be on equal footing with the other diets. The other graph clearly shows that as more arable land is used to feed people instead of animals, all three vegetarian options tie for first place.