Technically, in real life, every at-bat should be considered independent from the other. In other words, though we assume players/teams will come back down to earth after some incredibly hot streak, nothing says they absolutely have to. There is
some probability of successfully landing the ball in a safe spot past infielders and in between outfielders when you make contact with the ball, which is the most reflective thing of that law of averages... but that's not to say you can't get it there every time. The only thing that really changes that is how a team figures you out and plays against your hot streak.
So in a way, it's tough to say whether that player should become
worse as time goes on, or
better, or not change at all. I think there should be some variability in player numbers and that they shouldn't be so specific, but that's just me.
Just to go back to your Matt Kemp example, I want to say that future at-bats are independent in the sense that you can't just say "chances are" they won't do something. However, if you are able to
look back at something, then maybe you are able to form a better opinion on something. In other words, if Matt Kemp went 2-for-5 the night before and you are watching the archived version of that game (where you know that he went 2-for-5), and he singles in his first at-bat, you're going to say "chances are he won't get a hit his next at-bat," only because we already know that he only gets one hit in his next four at-bats. However, we can't predict that he won't do something.