The simplicity has nothing to do with it. For example, if you happen to be a very introverted person, then the social awkwardness of working in front of someone you don't know is enough to disrupt all your thinking abilities significantly. You're not only worried about the technical parts of the problem, but you are also worried about things like, "Oh god, I typed that semicolon in the wrong spot. This interviewer is going to think I'm an idiot..." and it's incredibly distracting.
This is a normal mode of self-conscious thought for a huge number of people who are otherwise excellent at their jobs and perfectly functional (e.g. introverts).
Plus, after you've worked a job for a while and know your team, know what's expected of you, and can get in the zone with some privacy, then these thoughts no longer really affect you. So they only play a significant role during an interview specifically because the interview is not at all a realistic representation of what working will be like.
I find it incredibly disturbing that you see fit to infer anything from the author's so-called "inability" to program "under stress." When someone makes that kind of inference, I think it says a lot more about their dysfunction as a colleague than it says about the other person's programming skill. Even if this problem had been Fizz Buzz or swapping two numbers, it's so, so far from simulating what a real work social situation would be like that it renders it useless for assessing the candidate, no matter if they answer it perfectly or fail completely. It's just a useless scenario.
