Because memory can be filled with random data, so that's a case that needs to be handled if a computer is to "perfectly simulate itself."
>> That's easy to disprove. A calculator isn't sophisticated enough to simulate itself, but a modern computer can perfectly simulate itself.
> Today we can run entire Linux in a web page and we can surely run a simulation of IBM PC.
You said "a modern computer can perfectly simulate itself." If you think "running Linux in a web page" is an example of that, I don't think you understood your initial statement. A modern computer can simulate a less-capable version of itself, but it cannot "perfectly simulate itself," as I showed in my comment.
It's not a good look to come in with an "easy disproof" that itself contains a pretty obvious falsehood. Also I'd wager your implicit statement that the OP's "to understand" is encompassed by your "to simulate" would be generally considered pretty controversial, at a minimum.
What you say is pretty obviously false: a computer with finite memory cannot simulate itself in a general case because that would require more memory than it actually has or a compression algorithm that's effective on random data, because its memory would need to contain the emulation program plus a full memory image. Computers with infinite memory do not exist and random data is not compressible, therefore modern computers cannot perfectly simulate themselves. QED.