Preamble you are allowed to skip:
My favorite games work like this, when they are GMed in such a way as to be fun to me:
-there is a problem
-the PC attempts to solve the problem
-because both the problem and the attempt to solve it are couched in terms of action in an imaginary world (i.e. the problem is "get the ruby away from the gnome" rather than "get three O's in a row before your opponent gets three Xs") the tactic used to solve the problem (successful or otherwise) creates a new problem (or set of problems) (i.e. now everybody wants your ruby.)
-the process starts over
-when this happens many times in a row it creates a little neverending story about somebody who solves problems or dies trying
(You may not like this, or prioritize it. Great. That's ok. I accept your right to be interested in something I'm not.)
Point here being: As a DM, I sit around dreaming up interesting problems to solve. The mechanics help me define the parameters of the problems and the things PCs can use to solve them.
To me, just as the point of combat and action mechanics is to facilitate the creation and in-game use of interesting dungeon puzzles and similar tests of tactical wit, the point of interesting social mechanics would be to facilitate the creation and in-game use of interesting social puzzles.
Putting players in a position to figure out how to think outside the box to disembowel a fully-statted owlbear before a fully-statted lava trap kills them is, for me, fair and fun and (a major) part of the game. However, it takes up a huge chunk of game time and if the players are not into this sort of thing, they will not like the game.
In order to justify social mechanics being as complex as combat ones, putting players in a position to figure out how to think outside the box to outwit or charm an NPC should be the same kind of challenge, but I seriously can't be fucked to think about NPCs for as long as I can to think about monsters and traps. At least not in terms of their personalities.
1- I'm not that interested in my NPCs until and unless they get some track record in the game (after all, I can write fiction whenever I want, and do. What makes them interesting isn't what they did, it's that action by the dice or PCs made them do what they did.),
2-If I don't get to use a well-designed trap or monster in an adventure, I can always have a similar one (perhaps re-skinned) somewhere else, whereas an NPC implies a whole gameworld social situation and the PCs don't want to interact with that situation then moving it around so they have to do it anyway is difficult to do without railroading,
3-Designing really good NPC interactions means much of the time at the table will be given over to NPC interactions, which isn't mostly why I'm here. I get enough of that reading books and comics, watching movies and writing books and comics and being in movies. I can have a pretend conversation with my roommate in the living room any day of the week (and sometimes do) whereas I can only put my mind to the question of how to get a gelatinous cube to absorb a beholder so I can immobilize it long enough to drop a boulder on it when I play D&D.
YOUR MILEAGE WILL VARY.
I KNOW YOU ARE NOT ME AND LIKE DIFFERENT THINGS. OK? GOOD.
But...it's about the journey and not the destination 'cause D&D the way I play is picaresque and therefore anything that happens along the way that's fun for the players and me and doesn't sap the possibilities of future fun is acceptable--any mechanic, any detail, any goal, any npc interaction, any bet, etc.
Therefore although I wouldn't want a social mechanic to be part of the 'bones' of a system I was playing in, I like the idea of an (always) optional gravy-on-top social interaction mechanic, so that social interaction can be detailed if the situation makes that seem like a good idea at the time to the players.
Using the social interaction mechanic instead of just talking to a guy would be like a PC deciding to play a bar game in a pub--they don't have to do it, but its effects are still binding on the fiction of the game if they decide to go that way and can have major effects.
Also, I think it would have to emphasize player skill over PC skill, and not in some way that just says "the player has to role-play his PC convincingly and skillfully", since if they just wanna do that then they don't need the mechanic, plus, like I said: we get enough of that at work.
Extremely important: the mechanic will require rolling on some ridiculous d100 table. The goal of the results would not be, for the most part, to simulate the statistical realities or storytelling possibilities of human interaction so much as to force the PC to roleplay in entertaining ways which exaggerate the monopoly-with-squatters effect.
Gygaxian Social Interaction Table
(roll below one time+once per point of charisma over 15 and pick the one the player likes best, if PC charisma is under under 6, roll one time + once per point of charisma under 6 and the DM picks)
(in most cases, these 'challenges' won't be secret, the player will know this is whats/he has to do...)
1-NPC talks fast and interrupts people. If the player can say what s/he wants in less than d12+6 words s/he gets it.
2-NPC talks fast and interrupts people. If the player can say what s/he wants in less than 2d4 seconds s/he gets it.
3-Conversation reveals NPC is obsessed with (conquest/unity/weasels, etc.). If player can work the word "weasel" (or "conquest" or "unity", or etc.) into every sentence of his/her request, s/he gets it.
(now I have to go take a shower, and I am turning the remaining results over to Gygaxian Democracy...)
(IMPORTANT: Remember the challenges must be things that a player could conceivably fail to do properly, not just things it'll be fun to see them do.)