in the event a honest guy tries to reference a legit tip and the attacker's tip, he'll detect the contradiction and won't do it. Therefore, the attacker's subtangle will be abandoned.
This requires not just an honest guy, but a diligent one as well.
The risk is that honest guys will be lazy and rely on others to go far back in history to check all tx for double spending.
The lazy guys risk that their tx's will be abandoned, because the majority of the nodes won't reference them.
That is precisely how I would have answered. It is quite clear that everyone has a strong incentive to be on a correct branch, else any time down stream someone can broadcast a notice that a branch is incongruent then that branch gets abandoned.
But doesn't this mean that there is a great incentive to not include tips in your branch, because these don't yet have enough veracity to be sure they won't end up being a double-spend. In your system there is often no way to prove which of the double-spends were first, so they both are invalid.
Seems to me no one has an incentive to lengthen instead of broaden the tree. But I haven't absorbed the white paper. Did you address that?
Yes. As mentioned somewhere above on this page (or maybe on the previous one), the (default) referencing algorithm works in such a way that it prefers tips with bigger height. So, if you're too lazy and reference some very old tx's, you take the risk that your tx won't be referenced by others.