A lot has been written in OSR circles about the siren's song of realism in role playing games. Pursuit of realism has led to some wildly complicated game systems, pages and pages of tables and subtables, systems and sub-systems bolted on willy-nilly, until you need an accounting degree just to keep track of a simple scene or combat. I think realism does have its place, and it is a good and useful thing when it remains in that place. Maintaining its proper place in relation to other factors of game design is key.
First off, it seems to me that people often conflate realism with level of detail, what I might call granularity or resolution. Resolving a combat with a single die roll is no less "realistic" than resolving it through a complex system involving many die rolls to represent actions within the combat, any more than a picture of Earth from space is less realistic than a 100-page coffee table book filled with pictures of people and places on Earth. They're simply different perspectives with different levels of resolution.
Secondly, I think that often the pursuit of supposed realism is actually driven by a misidentified desire not for realism, but for more choices in game play. A DM adopting a "realistic" system with hit location charts and rules to handicap characters according to their specific wounds may actually be intrigued by the role playing and tactical possibilities of playing wounded characters, but with his or her conscious focus on realism rather than on the true goal of providing interesting choices in play, the result may be a needless and frustrating quagmire of complexity. It may even be destructive and limiting to player choice, rather than expansive.
With those thoughts in mind, here's my take on the proper relationship between complexity, realism, and choice:
The goal of rules in a role playing game is to provide a framework in which players can make choices. The more interesting the choices it imposes or facilitates, the more valuable is a rule or sub-system to the game. Of course, what's interesting is a subjective matter, so the preferences of the players will tend to determine which activities warrant detailed "hi-res" systems and which can be glossed over with a single "low-res" die roll or ruling.
Greater complexity allows for more choices. However, complexity exerts resistance to the flow of the game, much like friction between moving parts of a complex machine. Too much complexity can offset the value of interesting choices provided by a rule or sub-system.
The purpose of adding complexity to a game should always be to increase the range of interesting choices, not to increase "realism." (The two are not mutually exclusive, but the focus must always be on the former.) Thinking of interesting choices as benefits and complexity as cost, the economic law of diminishing marginal returns clearly applies. At some point, the enjoyment gained from the choices provided by a new rule is going to be outweighed by the disutility of having to remember and apply it. This applies both to any given rule or sub-system individually and to their cumulative effect on the game as a whole.
Realism applies in a negative sense, not as a driving force behind system design but as a post hoc litmus test. Rules and systems should strive not to slavishly model reality, but simply to avoid clashing with it so blatantly as to harm players' immersion and willing suspension of disbelief and foil their reasonable expectations of how their actions should be able to affect elements of the game world. Magic and other fantastical elements need only be internally consistent,
but anything that mirrors the real world, such as armed combat, must not violate players' basic understanding of real world physical laws. Sometimes this is simply a matter of properly "fluffing" the rule; e.g. making clear that character hit points are the ability to avoid taking real bodily harm, not the capacity to absorb direct sword strikes.
In general, then, the most valuable and useful rules are those that effectively provide for the most interesting choices with a minimum of complexity and without flagrantly violating players' conception of reality.