The death of a player character is something every GM must face from time to time. Though the role of game master is not supposed to be strictly adversarial, the real risk of character demise is an important contributor to the excitement of the game. However, under certain circumstances, PC death might not be the most appropriate outcome, but neither is letting the players completely off the hook.
A past GM of mine introduced me to the tactic of a fate "more interesting" than death. This technique is useful in the event of the otherwise "accidental" demise of a character (i.e. one that occurs out of the blue, perhaps due to bad rolls, during some activity that is otherwise inconsequential to the plot.) For example, a dwarven fighter I played in this GM's campaign was attempting to lead a pack horse across a makeshift bridge over a deep chasm. The GM had me make a check to see how well I coaxed the animal across. I rolled a 1 (critical fail). The horse balked and slipped off the bridge. She then had me roll a balance check to see if my character maintained his footing. Another 1 (critical fail again!) Based on a strict interpretation of the worst possible two rolls in a row, my character should have been dwarf pate on the bottom of the gorge. Done in during an otherwise routine skill check crossing a bridge.
The GM, however, decided to take a more creative approach to this critical failure. Rather than killing me off, she had my character land hard, taking damage on a ledge partway down the cliff side. This set up a spontaneous encounter wherein the other players had to figure out a means for extracting me. The GM also added a caveat to my character's miraculous survival. She declared that the experience had so traumatized him that he now carried a permanent and crippling fear of heights. Almost as disabling to an adventurer as an arrow to the knee... but also an excellent catalyst for future role playing.
A couple weeks ago, my group had its first real encounter with player death. Others had been brought to a "dying" state, but this was the first time a strict interpretation of the rules should have meant "dead". Unlike my dwarf's experience, this death occurred in the middle of a major fight. So, why didn't I go through with it? Well, in this case, the consequences were largely out of the control of the person playing the should-be-deceased. His character had been immobilized by a spell and one of the enemies had put a knife to the PC's throat initiating a bit of a standoff. One of the other players had a chance to act ahead of the neck-stabber, but chose to attack her current target rather than responding to the threat. Neck-stabby stabbed and the immobilized player failed the check that would have saved his character. In this case, the situation felt a bit like this scene from Indiana Jones and the Last Crusade.
Though the knight declares, "he chose poorly" after Donovan gets all old and stuff, he didn't really choose at all. He let Ilsa choose for him (I know, handing off the decision is a poor choice in and of itself, shut up!) The point is, Donovan's fate, like that of my player's character was out of his control. So, rather than let the PC die due to the choice of of a fellow player not to intervene, I decided to invoke the fate more interesting than death. I dropped his character to a single hit point above completely dead (essentially bleeding out). This gave his companions a second chance to help. However, I also decided that the scar on the character's neck would remain, despite magical healing... a tell-tale mark branding a roguish character who often relies on disguise. If my fantasy world had a stock market, I would invest in scarves.
If a similar situation came up in the future, I'm not sure if I would take the same route, or if I would make the players burn some time, dough and xp getting their friend resurrected. After all, that's part of the fun, right?