Not true! It implies we might find performance optimizations, especially at the lowest level. Lazy loading, caching, pointers to constants, that sort of thing. It also doesn't discard Occam's razor. We actually have examples of simulated worlds (physics engines in games), so we know they are possible, unlike the flying spaghetti monster.
Nah, as other have noted, no simulation could have a 1-1 relationship between data humans observe and data in a physical device that exists in a world congruent to what humans observe - because there aren't enough atoms in the reachable universe for this. So such simulation either compresses the actions it simulates using higher level constructs or its happening in some universe congruent to the world we're in. Any such machine is going be a product of a future we don't know about yet and so it's constraints could be wildly different. Moreover, since the standard assumption of this simulation foolishness is that future humans or future post-humans want to learn about their ancestors, one can naturally assume you mechanisms that compensate for any "glitches" that might otherwise be obvious. Which just adds to my original claim.
The issue is that you can only learn whether you're in a simulation if the simulator allows you to do so. Otherwise, the moment that you discover a performance optimization, the simulator could just pause the simulation, delete the discovery from your mind, and resume.