Bostrom assumes all simulated minds will be "of a similar sort." Another assumption that "minds" are well understood.
Bostrom assumes we will never have the means to glean any information related to whether or not we are in a simulation. Indeed, compared to the unprecedented assumptions needed to admit posthumanity (computers the size of planets, for one!), the assumption that we can determine whether we are in a simulation seems paltry by comparison.
Bostrom's indifference argument is unfounded. In particular, say I only give his proof a 1/N chance of being correct for a very large N. Then by his indifference principle I _still_ must assign an overwhelming probability that I am in a simulation. This doesn't help in that I cannot be 100% certain his argument is wrong since he does not define things rigorously enough for a proof, nor can he possibly do so because of our lack of understanding about human minds or the future of computation. So to be a "rational" being (I hate this term, because it's also undefined), I can't disagree with his argument and conclude the opposite of what he concludes.
Since these are post-humans, they understand the subtle interconnected algorithms implemented by human brains, and so can in principle bypass the need for low-level simulation. But even if that were not so:
> In order to simulate a human brain, you need to simulate physics. Not a mathematical model of physics, but the actual true process that governs nature.
You seem to be drawing some implicit distinction you find meaningful between "real" and "simulated", but this distinction isn't explained, and doesn't seem meaningful in any case.
> In order to do that, you need to (1) understand the true laws of physics and
The laws of physics needed for such simulations are sufficiently understood by post-humans, ie. for all physical observations simulated ancestors might make, posthumans understand physics far beyond that. So even without a perfect understanding of physics, they merely need to replicate the physics understood by their simulated ancestors.
> (2) the process needs to be simulatable in polynomial time (at the very least).
I don't see why. We already run non-polynomial simulations all the time (such as to solve fluid dynamics for aerodynamics, or climate models). As long as the results are expected to be useful, we will run them. This is actually a core premise driving Bostrom's argument: that ancestor simulations will be useful to future social sciences and such.
But all of this is neither here nor there, because Bostrom's argument is exhaustive. If there are legit reasons why post-humans cannot run such simulations, and there very well might be, he already covers this possibility as "we will go extinct before reaching the posthuman stage".
Like I said, Bostrom's argument is solid and there is no wiggle room. You get to choose one of the possibilities, for properly articulated reasons, and no more.