Something that seems to frequently come up when discussing the design of a game system is whether or not some aspect of that system adheres to reality. Or, more precisely, whether the outcomes of that system accurately simulate the results that the person making the argument expects, based on their particular interpretation of reality.
Generally, these arguments come from players, or from non-designers, or less experienced designers, and will take the form of, "But XXXX isn't realistic!" or "Realistically, YYYY should happen instead". And, frequently, experienced game designers will turn around and say "Who cares?" and merrily go on their way designing an "unrealistic" system.
I wanted to give a quick explanation of why this is, explain what role I see realism as having in game design, and then provide a bit of a defense of "realism" as it relates to something I call the "responsibility of expectations" that is placed on any game design.
First - game designers primarily ignore or devalue realism because their primary goal is rarely to construct an accurate simulation of a real world. At their core, game systems are sets of rules that encourage and discourage, or reward and punish, certain choices within the game versus others. In establishing both what choices players can take, as well as the rules effects of those choices, designers are generally attempting to create a system which a) avoids "dominance" b) provides opportunities for players to differentiate their strategies based on the decisions they take, and c) give players opportunities to make "good plays" in response to opponents (be they computer or human controlled).
As a very basic example, take armor in Fallout: New Vegas. Wearing Light Armor has strengths (you move faster) and weaknesses (you absorb less damage). Wearing heavy armor has the opposite effects. Heavy armor does not dominate Light Armor, even though it performs better at its core functionality, because there are cases in which moving faster is more important than absorbing damage.
For a more complex example, because the game uses subtractive DT, low DAM weapons are disproportionately affected by DT. So, for instance, if a creature has 5 DT, going from a 6 DAM weapon to a 7 DAM weapon doubles your damage (your actual damage goes from 1 to 2). Whereas, if that same creature had 0 DT, going from a 6 to a 7 DAM weapon is only a 1/6 increase. This is an example of how you can encourage the player to make "smart plays" - when fighting an enemy in light or no armor, you are encouraged to use your highest DPS weapon regardless of its DAM. Whereas, when fighting an enemy in heavy armor, you want to select a weapon with enough DAM to significantly overcome its DT while still having enough DPS to deal substantial damage over time. When combined with the other properties of weapons (range, rate of fire, spread, etc.) you end up with an interesting matrix of choices in which players are encouraged to find the optimal weapon for any given situation. No weapon is dominant, players can select from a group of weapons they like based on their own personal playstyle, and there are opportunities for players to maximize their effectiveness through smart play.
These are the core goals of most game designs. Many games have other additional goals, and simulating reality is sometimes one of those goals. But when you evaluate a system from the perspective of someone trying to create as good a game as possible, these goals are paramount.
The problem with realism is that, in reality, there are strongly dominant options. Catch 22's exist. I can't imagine ever wanting to bring a knife to a gunfight. If I was going to be venturing into the Mojave wasteland, you bet your ass I'd want to only wear the heaviest power armor I could find and only use the biggest gun I had... plus I wouldn't have an invisible backpack full of other weapons I get to choose to optimize my damage in other situations.
So, often, the goals of realism and the goals of quality game design conflict, and in almost all cases realism is cast aside.
That explains why realism is so often ignored. However, there are two cases where realism is important - first, in creating a sense of "verisimilitude", and second, in dealing with the responsibility of expectations.
Verisimilitude is a term that, like "truthiness", I'll use to mean the extent to which something "feels realistic", even if it is not. For instance, again, I'll use eating, drinking and sleep deprivation, in F:NV's Hardcore Mode (if Hardcore mode is off, these features are not present). These may not actually realistically simulate the mechanics by which a person becomes starved, dehydrated or sleep deprived. The rates, and effects, are almost certainly not identical to what you'd encounter in real-life. However, the fact that I am thinking about food and water in the wasteland, and the fact that I am happy whenever I find a delicious, delicious fresh barrel cactus fruit, gives my journey through the wasteland a taste of reality. Not realism, but "truthiness", or vertisimilitude.
Note that these aren't really mechanics that work as I described above. It is a strongly dominant option to have water, and to drink it when you get thirsty. It's not really a valid playstyle to be "the guy that never drinks water". But, in this case, it is a small enough part of the system and has enough of a positive effect on verisimilitude that it's worth having the system, IMO.
Finally, when players approach a game, they come both with their own personal baggage/knowledge, and they form opinions on things in your game before they ever interact with the actual gameplay of those things.
As two examples, I'll use Civilization and Power Armor. First - in Civilization, one of the most common complaints has been the situation where 1000 spearman defeats a tank. It seems ridiculous that something like that would even be possible. Yet, in the game rules, it is clearly an outcome that is possible, and tailoring the rules to avoid that outcome could have deleterious effects on the actual gameplay, in which case it may not be worth it. In fact, allowing lower-tech units beat higher-tech units helps avoid over-rewarding players that fast-tech and start off isolated - if tech was as dominating in Civ as it is in the real world, the game would not be much fun, even though it would be more realistic.
However, players do come with that expectation, and that is something you have to deal with - so the ideal solution would be one in which the balance of the game isn't ruined but spearmen can't, in fact, beat tanks, because players come into the game with expectations and preconceived notions, and it's generally a bad idea to violate them (unless you're trying to make a point in violating them, which is maybe a bit too post-modern for a video game).
Next, let's take the example of Power Armor in F3. None of us knows how Power Armor works in "reality". So, our preconceived notions aren't about the actual function of the armor, but instead the weight that it's given in the art and lore of the world. When, in the intro movie to F3, the camera trucks out to reveal the fully power-armored soldier, you can't help but think "Wow, that guy is a badass. I bet that armor is awesome!" If, in the game, you then found that armor, and it was barely better than leather armor, you would be disappointed because your expectations had been violated - the game art wrote a check that the mechanics couldn't cash - and that's a drawback even if the game works perfectly well as a game that way! This is something that Rob Pardo talked about at the last GDC - it's important that you respect the fantasy in your game and not violate it through the game mechanics.