Hey! It's RPG theory!
At its most basic level GNS is fine: Any given decision made in an RPG can be said to be primarily made based on a single set of criteria. GNS identifies three criteria (or motivations, if you wish) for making these decisions. Narrativism means that the decision was made to address a theme. Gamism means that the decision was made to "step on up", to compete (it's actually a bit more complex than simple competition, but that's okay). Simulationism is, well, technically it's about fidelity to the source material and plausibility, but it's actually used as a sort of catch-all.
This is all fine and good, and to some degree useful. The next step is fairly logical, let's track trends in play! We'll see which of these three (arbitrary) criteria a player uses most often, and that will help us to understand why he plays!
This can be useful, but there are some significant problems:
1. The categories are arbitrary. Based on observation it was decided that all decisions that occur within roleplaying can be categorized as one of these. Now, it's important to note that GNS theory as it currently existed grew up on The Forge. To be fair they're a good group of guys overall, but they have their priorities. For the most part they were all about addressing Theme, and for whatever reason were secondarily concerned with Competition. The result was that anything that didn't clearly fall within those two categories kind of got shunted into "Simulationism". No big deal really, but I'm not sure if anyone noticed it happening or really picked up on the fact that the result was a terribly snarled beast the category became...
2a. As Vincent Baker notes, despite the efforts of the "old guard" to avoid it people lost sight of the fact that GNS can (at best) only identify the reason for a specific decision and can be used to look at trends in decision making. The result was that most people use them to pigeonhole people and treat the categories as opposing values, neither of which are intended uses of the theory.
2b. And in all honesty, to some degree it's the "old guard's" fault. Don't get me wrong, I think we're much better off with GNS theory and its legacy than we would be without it, but some mistakes were made. The big problem here is this intense focus on mechanical "facilitation" of a given Creative Agenda. There's this pervasive sense (it may even be explicitly stated somewhere, but I don't know) that you can really only facilitate a single Creative Agenda well in any given game. This, I think, was a major contributor to the idea that these are discreet categories of play that don't mix well.
3a. While at its base GNS is very logical, it sort of falls apart in the real world. Yes, it is true that any given decision is made primarily for a specific reason, but it turns out that it's not always (or even often) clear which reason is primary. My current favorite example of this is Capes which focuses on competing to address theme. It's easy to see what Creative Agenda is at work when a situation arises in which all three are opposed, but much (most?) of the time at least two, if not all three, will have the same results. I don't really have a solution to this, but it makes the whole GNS thing a bit artificial feeling.
A lot of this problem arises from the simple difficulty of analyzing these things in yourself and in others. Do I choose to address theme primarily because I think theme is important, because I want to impress others with my insight into that theme, because it "just makes sense" to address theme? And, even worse, this changes all the time. A game that has been about addressing theme for a while often starts to show elements of competing to address theme, and addressing theme because "that's what this game is about."
3b. The fact that a decision can easily fall within all three of the Creative Agendas gives lie to the idea that you can only facilitate one of them well. For example the aforementioned Capes makes it a competition of addressing theme. I could envision a game that makes a competition of maintaining in-game plausibility in a world/genre/whatever that is all about addressing theme...
Let's bring it home...
GNS was great. It really helped get across the idea that people play for different reasons, but more than that it got across that sometimes these reasons are in so much conflict that you can't play. GNS grew into something more complex, people tried to use it for things that it isn't really suited to handle. It was used to artificially separate play (see Vincent's GNS diagram for an example of this, Competition and Theme aren't mutually exclusive as are implied there).
Do I think GNS is useful? Heck yeah! But only if we understand its limitations:
1. It is arbitrary and not rigorous. There are reasons to play other than the three defined.
2. It can not be used to categorize things. We can talk about individual decisions, and (for what little it's worth) trends, as long as we keep number 1 in mind.
3. The things that GNS splits up may be technically mutually exclusive, but you can't always (or even often, I believe) identify which is at work. I'm pretty sure that the result is that "facilitation" of a given GNS Creative Agenda is a "yes and" not an "either or" proposition. You can facilitate Competition, and Theme, and other stuff all at the same time if you want.
So lets stop talking about GNS and start talking about making a game facilitate Competition, or Theme, or Verisimilitude, or Plausibility, or whatever else. And then lets talk about how to make a game facilitate multiple of these.