> In general, you can't simulate physics on physics faster than physics
You assume that the physics of the simulation are equivalent to the physics outside it. We simulate simpler environments than our own all the time.
Maybe the constraints of our universe are the way they are because it reflects the tradeoffs required to be able to simulate it given the conditions the simulation runs in.
Maybe, but that's not interesting. It is possible that the physics inside are different from the physics outside, but that tells us absolutely nothing of value about whether we are in a simulation or not.
If we do assume that an observer cannot differentiate the simulation from the outside world, then we can make an interesting argument: if the rules of physics in some universe entails that this universe will simulate many other universes with the same apparent rules, then an observer of these rules is probably simulated regardless of what the root universe happens to be. This is the only way you can ground the probability calculation. A lot of things are possible. Not many things are probable.
For an observer to be unable to differentiate the simulation from the outside world if put in the simulation is a very low bar, and the one that matters.
It's a low bar because you don't need to presume the simulated entity gets the benefit of the passage of time. For it to be likely for us to be in a simulation, it's sufficient for it to be likely for the total runtime of "simulated us" to be higher than "real us", and our only knowledge of the passage of time are the momentary memories and sensory inputs right now, and we can't validate their truthfulness.
In other words: you don't need to simulate any physics to prevent a mind from differentiating. You just need to limit the scope of your simulation to a sufficiently narrow slice of time.
Personally, I tend to think that if simulations become widespread, test runs, debugging and development, research runs, or application specific runs of small slices are likely to far outpace full scale simulations in runtime, and so if simulations are probable, I'd be inclined to think we're most likely in a (very) short-lived, scope-limited simulation in a horrible groundhog day like setting.
> In other words: you don't need to simulate any physics to prevent a mind from differentiating.
But if your simulator is not simulating physics... what is it simulating? The mind? If you are simulating a human brain, that IS a physics simulation, and it is probably on the harder side of the scale, so we cannot readily assume it's going to be easy or fast. And if you are running some different kind of mind based on an abstract neural network or something like that, well, it's not a simulated mind, it's just a mind.
You assume that the physics of the simulation are equivalent to the physics outside it. We simulate simpler environments than our own all the time.
Maybe the constraints of our universe are the way they are because it reflects the tradeoffs required to be able to simulate it given the conditions the simulation runs in.