You are right. Even more, some percent of double-spendings can be tolerated if we save a lot of resources by allowing this. Every day new counterfeit dollars appear in the streets but Earth doesn't stop spinning. If we accept that it's fine for FED to print more money, why a skilled guy from Harlem can't do the same? Hell, I would prefer the latter, at least he can't print that many banknotes.
The problem is without Consistency I expect the DAG to diverge into a chaos of disagreement. You are relying on participants using clients that adhere to the mathematical model you want them to use when choosing which tree branch to append their transactions to (and which acceptance model to use for declaring a transaction is probabilistically confirmed), but given the inconsistency that will arise and the game theories thereof, I don't see a snowball's chance in hell of the thing not blowing up unless you are able to maintain control over what participants do, and then it is no longer decentralized.
My point is there is no equilibrium of just a low rate of double-spends, but rather divergence. I haven't shown this formally (as in a math proof with equations) but I can already see it conceptually.