Not necessarily spurious. It could be, that, say, local poverty correlates to both people eating margarine instead of butter and people getting divorced. In this case, I think that's unlikely, but it's a possibility. The more correlations you look for, the more you have to "discount" the p-value you get from them, which is a major problem in, say, psychology research. In the case of marijuana and drug overdoses, it seems like something fairly obvious to look for, so I wouldn't imagine that it's just a result of people going through all possible correlations between their variables fishing for a p-value, but the article I saw about that didn't even quote their p-value, so while I can say that I wouldn't discount that one too much, I don't know if it's good enough to take seriously.