Perhaps people with naturally smaller reward regions are more likely to become heavy users. With a smaller reward center it would take heavy use to achieve the same dopamine levels a neurotypical user would get from a small dose. The paper itself cites the study of 12-year-olds you mention and says that it is the same abnormally small OFC volume that is the predictive factor for MJ use later on.

They claim they have demonstrated a causal link, but do not state how they controlled for self-selection among applicants.