hi,

this is an fgs analysis. Here yo can see that with fgs we should shove this.
HRCFGS3A

here I adjusted the calling ranges for my shove accordingly. I expect them to be looser than nash. This makes it a fold. The ranges a really tight expectations. In reality they will cal looser. Which makes this one even more a fold.
HRCFGS3B

Now I change the shoving ranges of the remaining players. These guy never ever shove this tight or this loose into each other. Here as you can see because they get it in a too small fraction of time we can profitably get it in.
According to this we should estimate the likelyhood of bb and sb getting it in even in a don shove if the dont.
HRCFGS3C

You may not also if you change the ranges slightly here is differs from clearly shoving any 2 and folding almost anything. In a non don enviroment its an easy shove but here where I want to minimize my variance its much closer.

So by definition you can go with your first one and not be exploited, but you are correct that if they start to call much wider than Nash, it hurts your expectation (but it hurts their own as well).

I struggle with this myself a lot and haven't seen it "solved" as much as discussed more "in theory". But I still wonder the ideal approach to games where players are so far off from Nash. Generally the further someone is from equilibrium the more exploitable they are, but here if they shove significantly tighter and call significantly wider, it pushes you towards an "exploit" of fold everything all the time, which is what some OMCs who are not winners do at these games.

It's not really backed by much, but I frequently just "split the difference". In spots where I know I should be shoving nearly everything, but vs. their adjusted call range I should shove nearly nothing, I often split the difference, kinda tailored to how loose they are in calling (or how likely they would be shoving into each other were I to just fold my pockets jacks). Variance won't be tamed.