My own thinking is that if you're going to include these mechanics at all (and there's an argument to be made that they add drama and tension to the die rolls), you need them to be a function of level. The absolute worst outcome of this in a vacuum is exemplified by D&D with crit fails and iterative attacks. Many crit fail systems have the crit hurt you, and don't scale either your chance of getting a crit fail or the consequences of one by level, so that a level 20 fighter is likelier to decapitate an ally than a level 1 fighter. If you're going to have crit fails at all, you need to use them as an element of design. By that, I mean that you can't just tack them on - they need to serve a purpose other than "occasional hilarity", and it needs to be more specific than "adds drama".
Personally, I prefer to scale them so that you are less likely to get a crit fail as your skill improves. The purpose this shows, then, is that you are less reliant on luck because you're sufficiently badass that your own skill is the core of what you need. It doesn't matter so much that maybe there's inconvenient footing that might trip up a lesser swordsman, or that you're fighting in close quarters with allies you don't want to hurt - you've got the skill to handle that. Critical failures then become a way of enhancing the feeling of advancement from weak characters to those with greater control over their own destinies. For instance, if I put some kind of auto-miss or critical failure system into a game, I would supplement it with the following houserule:
"Whenever you make an attack roll, a high base attack bonus allows you to make extra rolls, from which you choose the best. You roll 1 die for every attack granted you by virtue of your base attack bonus (1 at BAB 5 or less, 2 at BAB less than 11 but greater than 5, and so on). Other sources of extra attacks don't grant you extra die rolls.
You threaten critical hits as normal, using the die that ultimately determines your attack roll. If all of your dice are natural 1s, then you suffer a critical failure."
I probably won't add such a houserule because table lookups for this sort of thing are clunky and annoying, but if I came up with a unified mechanic it might be useful for lending some particular feel to a particular campaign. It should also be noted that most of my games are electronic, so I'm disregarding the slowdown of rolling that many dice at once and finding the highest, since it's a nonissue in my play environment. I absolutely would not use critical failures at a meatspace game, because I don't think it's possible to make them serve a good purpose while also keeping play streamlined.
