All the previous discussion about fights also implies something that I figure I'll state outright - writing a generic combat system is a much bigger pain in the ass than writing one for a genre. So much so that most generic combat systems aren't - they are in fact implicit declarations of genre through the simple fact of how they handle things like survivability, "realism" and such.
Not to say they are unaware of this. Most provide a baseline with the expectation that it can be tweaked to handle other genre expectations (provided the game acknowledges genre expectations in the first place), but they need a starting point. So with that in mind, let's establish one.
My personal burden is that the two poles I am drawn towards in RPG combat are strongly opposed. The first is Swashbuckling, which is a bit over the top, but not as much so as Hong Kong action or Anime. The second is Rolemaster, where violence is dangerous and risky and always to be taken seriously. Those two things are hard to combine, but ideally we can pull off some sort of vinaigrette sort of emulsion.
This means the first thing to look at is stunting. This is sort of a broad question best phrased as "how tolerant is the system of players doing cool stuff that might be unrealistic or tactically inappropriate but which looks awesome?" This is a kind of important question because, practically speaking, swinging from a chandelier is a pretty awkward way to make an attack, but it looks pretty cool, so how should a game handle it?
The first option is the strict realism school, which would impose a penalty to the attack based on the difficulty of it, maybe call for an athletics roll to pull it off, and otherwise frown and tut-tut at the idea. In short, everything in the system would suggest that this is a terrible idea, and one you really shouldn't pursue unless absolutely necessary.
There's also a loose realism school which is not necessarily going to penalize such an attack so much as make it non-optimal. It might offer some small benefit (such as allowing the attack to be made with an athletics skill) for a serious tradeoff (you do unarmed damage, which probably sucks). This is something you can do, but it's rarely going to be the optimal thing to do.
At the far opposite end, you have the true stunting school (names thus for it's use in Exalted) where the character gets a _bonus_ on the attack based on how awesome it is, thus making colorful attacks desirable. This can get a little silly as players ham it up to get the bonuses, but it definitely supports the flashy.
Between those, there's a broad band of cinematic styles, ranging from abstract cinematic (like 4e's stunts, where the mechanical effects are sufficiently disconnected from the fiction to allow a lot of narrative flexibility within fairly strict mechanical interpretation. Also applies to many scene-based resolution games) to gritty cinematic (as in Feng Shui, where stunts are penalized, but that penalty is very small and bonuses are very high, which allows them but discourages their constant use).
This cinematic space is probably where I want to aim for, which is definitely on the swashbuckling end of things, but it highlights a few things. My real goal is the 1970's Three Musketeers movies, where fights have flair (and even humor), but also have a sense of danger to them despite reasonably limited bloodshed. That means I'm not looking for Hong King musketeers - it sucks in film and it's not what I want in a fight.
Part of this is tied to the cultural role of violence and death. It's important to remember that while we remember the duels and swordplay in the Three Musketeers, that was criminal activity. Duels are illegal because people killing each other is a terrible thing. It's a sin and a crime, and while you might try to outright murder the other guy on the battlefield or in certain brutal circumstances, those are edge cases. Or should be. The problem is that in any modern game, you can be certain that a player wants to be Wolverine (or the like), and such things quickly move towards the least common denominator. That's rough.
And with that, I find myself wrestling again. Is there a meaningful way to address the role of conflict in the game outside of the conflict rules? Setting design can speak to it certainly, but there's a reasonable case that they can be quite toothless in the face of a setting that says one thing and rules that say another.
Argh. Ok, I clearly need to just write some rules, then worry about this issues _after_ I do so. Doing so beforehand is proving utterly paralyzing.