The x - ˆx property is very easy to avoid when building a simulator. Most server-grade computers already use error correcting codes for their memory. Or, the simulator could just abort and restart at a recent checkpoint if an error is detected. It's possible to detect errors with arbitrarily low false-negative rate for a small additional cost of computing and storing checksums.

Nevertheless, it's an interesting observation that we can now easily do experiments that demonstrate correct behavior of logic to the 10^-15 level. If Descartes were looking for evidence of the fallibility of a daemon creating his sense data, it would have been hard to demonstrate better than 10^-3 or 10^-4.

You're right of course. Nevertheless there's a difference between saying "the simulating computer probably uses error-correcting codes or something" (speculation) vs. saying "an experiment suggests (same thing)" (science).

To borrow from Nick Bostrom: suppose we run two types of simulations. Important simulations and un-important simulations. For the important sims, we use error-correcting codes, we save checkpoint images, etc. For the unimportant sims, we don't do those things, in order to save money. This allows us to run far more unimportant sims than important sims. Thus, if someone is incarnated randomly in one of the sims, it's probably one of the cheap ones (just because there are more cheap sims than important sims, by basic economics). The point is just to show that it is possible for a philosopher to argue against error-correcting codes etc. Indeed, if we leave it to philosophers, we'll probably never make progress.

We need to appeal to the muse of science, that harsh mistress who serves us cold hard facts, every single one of which throws 50% of philosophers out into the darkness where there is wailing and gnashing of teeth :)