I've always had a simple critical hit system in my games - on a natural 20 you roll a d10 and on 8-0 your damage gets a multiplier tacked on.
I've also always had fumbles - on a natural 1 you roll a d6 and on 1 you fumble; there's a table to see what you've done to yourself and-or your allies. Bane effects and similar can increase your fumble chance.
The thing that bugs me about a lot of critical, and fumble, systems is their pure randomness. The above demonstrates that. Skill has no bearing - 1 in 120 of your swings will be a fumble, whether you are a commoner or a 20+ level warrior. That warrior will actually fumble more often, since he will get multiple attacks. Shouldn't skill reduce the likelihood of a botch job and enhance the potential for a more devastating hit?
If we have an army on the field, say 600 men to each side, trained warriors, but perhaps still somewhat green, does it "feel" right that 10 of them will drop their weapons or worse, while 18 will strike a critical hit, on just their first swing? Might be OK for the Critical - some hits simply strike more true. A Fumble every 4 minutes (10 rounds to the minute, 3 swings per round) doesn't feel like an expert and experienced warrior to me. It seems more like slapstick.
it has nothing to do with DMs wanting to "channel player behavior" and more to do with the idea that no matter how lucky you are (hit points in all editions being partly defined as luck) there's always a risk.
Actually, it has everything to do with channeling player behaviour, whether the DM set out to do so or failed to realize the impact. Greater risk of combat logically converts to players making greater efforts to minimize combat, or stack the deck in their favour, as few players want to make a new character every hour or two. Greater risk of character loss/shorter character lifespan also logically leads to reduced player investment in the character - why would I spend four hours putting my character together if he's only likely to last a game session or two? I'll just pull out an old character, change his stats and personality a bit, rename him and in we go (or just erase the roman numeral behind his name and replace it with the next one in sequence).
And keep in mind crits don't have to always be auto-kill; it could be as simple as saying that crossbow bolt got you for 18 points damage instead of 6.
This is key, to me. Why can't the results have varying severity and duration? A fumble could mean you stubbed your toe, missed your attack, recovered your footing but are off-balance and suffer a -2 AC penalty until your next round. A Critical could leave a painful wound imposing -1 to 4 on actions using that limb, with the penalty decreasing by one per day as the wound heals. A bad critical might break your wrist, rather than sever your hand it will heal, but it will take considerable time. New healing spells to deal with such wounds, or added benefits of existing spells, could also be used.
Applying "should be's" to casting magic spells is always questionable to me - it's pretty tough to objectively apply realism to magic, as magic is inherently not realistic.
A lot of it comes down to the game feel you want. If combat is very swingy, it's reasonable to expect some players will make every effort to avoid combat. Others will avoid making much investment in characters, and rush into combat hoping to get lucky. If not, change the roman numeral on the character sheet. If you get lucky, maybe some investment in this character, assuming his survival odds are now improved, might be warranted. If players can predict "the crossbows can't kill me", then action movie scenes are more likely. Which is desirable depends on the feel of the game you want, and a lot of that feel is player and character behaviour.
A GM who wants the players to respect those six crossbowmen so he implements critical hits to make it more likely they will. That impacts their attitude to all combat. I'm amazed how often a thread or discussion that starts with "my players never do Genre Behavior XYZ and it makes my game feel off" quickly reveals that, whether due to GM or game system, doing XYZ is disadvantageous. "The Good characters are bloodthirsty and never spare a defeated foe" is one classic example. We then discover the game gives full xp only if the enemy is killed. Discussing the GM's campaign, every foe the players have ever spared comes back to threaten and betray them in the future. Gee, it sure is puzzling why they never spare a defeated foe, isn't it?