I asked if there were any games where the PCs are not working together, and a few examples were suggested. Cold City is a great example of how a strong group structure can be used in a game. All the characters are part of a secret police force set up to deal with left-over Nazi occult weapons in post-war Berlin. You’ve got plot hooks and a reason to work together right there. Each PC is from a different country (a great roleplaying hook) and has a secret agenda (conflict! betrayal! plot complications!). There’s a Trust mechanic to bind it all together and bring it to the fore in every game session.
From what I can recall of Cold City, though, there’s no scope for the player vs player intrigue once Trust has been broken. As soon as the villainous Dmetri steals the zombie formula, betrays the rest of the group and flees back to Russia, that character is gone from the game. Once you leave the group structure, you’re out.
That restriction on player-vs-player action is common to a lot of games. You can betray and manipulate other players, but only so long as you do it within the party. A player character that’s in direct conflict with the party won’t stay in the game for more than a session or two, unless the GM is very indulgent. There are two major reasons for this.
First, handling time, which is the evil twin of spotlight time. The party concept lets the GM treat all the characters as a single unit – they’re all in one place, talking to the same NPC (I’m always amused in Call of Cthulhu games when half-a-dozen antiquarians and private detectives and occultists all simultaneously interview one traumatised witness), doing the same thing. When you split the party, you increase the handling time; when you split the party and some players need to have secret meetings with the GM, the handling time skyrockets.
I once ran a Legend of the Five Rings game with a lot of political intrigue; the first half or more of every session was taken up with private chats in the kitchen as the players connived against each other. I deliberately wrote rules into High Programmer, where you have to buy private conferences with your in-game resources (the justification is that there are so many spies and bugs in Alpha Complex that you have to take special precautions to speak without eavesdroppers, but really it’s just to keep the game moving).
The second problem with player-vs-player antics is that it’s potentially not fun – or, more accurately, it becomes immensely unfun when one player wins. A player is a fantastic adversary – the popularity of multi-player computer gaming is testament to that – as the opponent isn’t limited to ‘playing fair’ like a GM. The problem arises when one player gets the upper hand; even if he doesn’t eliminate the other player, it’s still in his best interest to restrict the other’s ability to act. In one D&D game years ago, the thief made a deal with a dragon that potentially endangered the rest of the party. When they found out, the wizard piled on dominate person and geas spells so thickly that the thief effectively became an NPC. The same freedom of action that makes the player opponent so interesting means the consequences for defeat are much more serious. Other players can screw you in ways GMs would never dream of.
Properly limited, it could be an interesting design space – you’d have to ensure that killing or imprisoning another PC is never the best course of action for a player, and come up with ways to handle secret action without bogging the game down. It’s very tricky, though, and I think the more traditional ‘conniving within the group is good, but overt conflict within the group is bad’ approach is better for most games.