I would like to be convinced that there is, actually, a real-world problem before thinking about potential solutions.

I'd like to see more analysis of the proposed selfish-mining algorithm at a particular share-of-network and gamma=0 (assume second-broadcast blocks always lose, to make the math easier). I can't reproduce the finding in the paper if I take into account the "opportunity cost" of working on more blocks in the private chain that might be orphaned instead of always simply extending the public chain, but it is very possible my little brain is missing something obvious.