To start off, "Realism" means that you're conforming your game mechanics to how you think reality works. Usually realism is good and necessary, but sometimes it'll get in the way of making an Awesome Game. What I'm going to do here is I'm going to try to explore that distinction.
To start with, is realism necessary? Yes. yes it is. Moving on... oh right, the reasons. To start with, realism allows your players to grasp on to something, which helps them understand the game. If I told you I was making a game where, all else being equal, the side with the fewest soldiers won a battle you'd be rightfully skeptical. Our experience with wars and the way the world works tells us that it doesn't happen that way, so I'd better have a darned good reason for running it that way in the game.
Furthermore, people expect a certain level of realism. They'll suspend disbelief on the points you tell them to, but otherwise they expect the world to act like, well, like they'd expect. My Grandpa watches reruns of Bonanza. One day we were watching one and he says to me "why does that Indian have Little Joe's horse?" Now he knows and I know and you know that it's just a TV show, that they've got constrained budgets and that when they need a bunch of mounted Indians they have to borrow some mounts from regular cast members who aren't on screen at the moment. But seeing the horse where it oughtn't be breaks the illusion and makes us unhappy.
Suspension of disbelief isn't a bad thing mind you. Far from it! But you've got to manage it well. People are going to accept all sorts of things that they know can't happen in real life, but you can't abuse that trust. You've always gotta make it sound plausible, and you've gotta be careful about contraditions in your internal logic. (Like that horse. No plot reason for it to be under that Indian.) Suspending disbelief doesn't hurt your realism, but breaking that suspension does.
Take Super Mario Bros. as an example. We understand that jumping on things heads is generally bad for the thing being jumped on. We understand that falling in pits of lava is generally fatal. We know that jumping on things with spikes or upwards facing jaws are probably going to hurt us more. And, while we might be a little fuzzy on why a hundred coins equal an extra life, we know that collecting coins is good, because we want to do so in the rest of our lives. All of these mechanics make sense from a realism perspective, even if they're expressed in a very unrealistic game.
Except for that 100 coins bit. Why does it give you an extra life? Well, this one is harder to justify based on our expectations of reality. I can tell you why it's in there, it makes for good gameplay. We want to collect the coins anyway, so they give us a bonus for collecting the coins. Collecting coins is fun, getting a bonus is fun, doing both is more fun and so the mechanic is added. Realism will be ignored when the result makes the game a better game to play. Come to think of it, who ever told you that you'd get 3+ tries at life, depending on how many mushrooms and coins you collected?
Thus you have two forces pulling in different directions when you're designing your game. Realism makes your game easier to understand, and furthermore people expect it. But fun often constrains you to make mechanics that aren't very realistic, for the sake of making the game playable. Take Evil Genius, for example. I was arguing in my previous post that the research system might be more realistic, but it was less fun. In that case it seems to me that they went too far towards the realism side of things. I, on the other hand, tend to err towards the other side of things; less realism, more action, which doesn't necessarily make it more fun.
Let's talk about the Terrible Secret of Space for a moment. I spent a while figuring out how to run all the orbital mechanics. I think I'm going to ignore almost entirely liftoff from planetary surface. Why the distinction? Because having planets and distances change dynamically over the course of the game gives an interesting battlefield, which changes from turn to turn. But dealing with liftoff from a planetary surface just adds complication to the game without much in the way of interesting mechanics, strategic decisions or so forth. So I spend time dealing with the one and nearly none dealing with the other.
But what does Realism have to do with all of that? I court realism with the movement mechanics (Remember, they're all based off of physics) because it gets me what I want; an interesting game. I couldn't make a variable board like that without realism because I wouldn't have a mental hook to hang it's movements on. But I'm avoiding realism with the liftoff questions because as far as I can tell it adds complications without adding anything of value.