So, anyway:
Permadeath.
My opinion is that how a game handles the end of a player's character is a design feature to enable the game to achieve its purpose. Of course the immediate purpose of a game is for the players to have fun. But there are a lot of different things that are fun, and those things include being entertained by having a story told to you, having a competion of skill against others, improving your skill with competition against yourself, having a social interaction with others, and telling a story. Different sets of games try to create fun in different ways.
If the purpose of your game is to entertain the player by telling a story, the player's character dying is an impediment that must be overcome. If you stop before the story is over, your game doesn't achieve its purpose. So, yes, if you have a cinematic game with a story arc and cutscenes and professional actors who are acting out a particular story -- then what you're selling is that story, and the fact that the player can cause the story to stop by carelessly "losing the game" allowing one of the major characters to die -- is just stupid. If you let that stop the story, your narrative arc won't reach completion.
If the purpose of your game is to be a competition of skills against other players, you don't even really have a concept of player death as something to handle. I mean, tic-tac-toe, chess, checkers, poker, etc. You can't "die", and there's no question about what to do when someone "loses before the game is over" because losing is, in fact, the definition of the game being over. Roguelike games that have a scoreboard used by many players function as competitions, to some extent, but that's not the primary use of roguelike games now.
I think one of the main purposes is to be a competition against yourself. I think a major part of the fun is to build your skills, so that's where the single-player scoreboard is meaningful. Doing better than I've done in the last ten games is important; getting to a new landmark in the game reinforces that I've learned to do something right. So yes, that's a me-against-the-game competition or a me-against-myself competition, and messing it around by pretending games don't end deprives me of feedback and deprives me of focus on the part of the game I need to work on. So I consider that a valid reason for permadeath, or at least a good reason not to lightly use the ability to avoid it.
There's another category of games whose purpose is not telling a story so much as it is creating stories, and these are roleplaying games. When your second-level wizard in D&D unthinkingly aims a fireball wand and sets off a fireball with a 20-foot radius in a room with a 20-foot diameter, while standing inside that room ... well, that means you roll up another character in order to continue playing the game. That second-level wizard's story is over, but the game it was part of - the campaign whose story is being created - is not. The difference here is that the content of the story is not a foregone conclusion. It is not written. You don't know when you started that wizard whether his story was going to be the novel of a powerful being who rose to challenge Gods and Monsters, or a cautionary short story about why you should be aware of your equipment's effects and take basic safety precautions. If you're cooperating in the creation of stories, you accept the story you get and move on to create a new one. And on a higher level, of course, there's the campaign story - the epic that that silly wizard's story was a small part of, as you mark the growth of your own skill and your ability to tell longer and more glorious stories in this medium progresses.
I know there's contention on this point, but I see roguelike games primarily as roleplaying games - in the sense of cooperative creation of stories, not telling a story. The end of a particular character doesn't present narrative problems, because there was no *other* ending that particular character needed to be destined for. If repeating content presents a problem, it just means your procedural content generation needs to be better to present a greater variety of story beginnings. Unlike the game that's produced to tell a particular story that required scenes to be filmed and plot points to be planned, you're working with procedural content - a system that's just as capable of telling the next story as it would be of continuing the current one.
So anyway; short version. If you see it as a game where you're telling the already-written story of the ultimate winner, and there is no other possible story for the ultimate winner, then players losing are simply an obstacle that must be overcome and permadeath makes no sense.
On the other hand, if it's a competition, or a game to build skill, or an excercise in creating stories, and you have already accepted that there are many, many, profoundly different stories that can be told about the ultimate winner and you have already accepted that there are many profoundly different stories that can be told about those who strove and failed.... then it is impermanent death that makes no sense.
So - of course I see roguelikes (or at least the ones I like) as having the last three purposes and not the first. And therefore character death as simply a legitimate ending of the game. Why would I mess up a perfectly good story (or competition, or lesson) that's over by bringing a character back?
