Think of the derivative of the red curve. It represents something like "for each marginal person who switched their behavior, how many total people would switch after counting the social effects of seeing that person's switch". If the slope is less than one, then small effects have even-smaller social effects and fizzle out without a significant change. If the slope is greater than one, then small effects compound, radically shifting the overall expression of support.
I think most of these are "secretly adaptive/reasonable" in certain contexts.
Fundamental Attribution Error: Reduces computational load when predicting the behavior of strangers in short interactions.
Conjunction Fallacy: It's harder to tell a complex lie without getting caught, so complexity is evidence for honesty.
This graph looks like it's just counting the fraction of services in the category rather than having anything to do with revenue.