Here I am not talking about computational power in this "absolute" sense but in the sense that a model A is fundamentally worse at doing a task than a model B if model A requires factorizing large prime numbers to solve it while model B doesn't.

So when I say "possible" or "computationally feasible" I mean that the orders of complexity of various tasks might be unnecessarily high without a hidden state. One can then go on to construct examples where this really makes the difference between "practically possible" and "practically impossible for any alien computer of the size of earth".