Your PCs must be the best-behaved party in all of D&D in that case, as they have clearly never:
- attacked an innkeep after being shortchanged
- captured a street urchin picking pockets
- started (or tried to stop) a town riot
- been involved in (or, as happened in my game, been the quarry of) a torch-and-pitchfork brigade
- participated in a bar brawl with the locals
- needed to neutralize a town guard in order to sneak in
- needed to neutralize some prison guards in order to reach someone and free him
- been involved on either side as a local farm gets raided
- etc.
All of these require at least some mechanical representation of the commoners involved
With the attacks against random inhabitants (urchins, innkeeps etc) I would be likely to make success automatic: if the players want their PCs to kill such people, they succeed. In my game, what would make that scene interesting is not its mechanical resolution, but its downstream consequences in the story.
As far as neutralising single or even small groups of guards is concerned, I have (and would) resolve this as a skill check or skill challenge: on a successful check, the guard in question is "minionised", and a power can then be used to kill him/her. If the check fails, then a "real" combat is required - the logic of this is that, on a failed check, the player doesn't get what s/he wants - the guard has the opportunity, within the mechanics, to try and flee for help.
Riots and brigades sound to me like swarm rules. I have run a couple of recent encounters with hobgoblin phalanxes, statting up the phalanxes as swarms (that had the ability to "kill" adjacent hobgoblin minions in order to heal their damage).
Bar fights are not well modeled by rules that assume violent confrontations where you're trying to kill someone.
Agreed.
Skill challenges based on the difficulty of what's being attempted. How well the Urchin knows the streets isn't in any statblock I'm aware of.
<snip>
Most of them require a represnetation of the mob or the city rather more than the individuals.
More agreement.
Yeah, yeah, skill challenges, blah blah. But you're basically having to come up with all of those on the fly or otherwise putting the situation into your favored mechanic and setting the difficulty. Why not use the mechanic that's already there and been there since the beginning? Combat rules.
In my own case? Because the combat rules (in 4e, at least) are for running interesting combats, not the cutting down of urchins by heavily armed and armoured warriors! As I indicated above, the only reason for moving to tactical resolution would be to give the NPC a chance to escape, thus introducing some extra complexity into the situation.
As to your comment about making up stuff on the fly - running a skill challenge on the fly is no different from statting up an urchin on the fly. That's what the game has encounter build rules for, to support this sort of stuff!
Failing to at least try to model the world - even to model the internal game world to itself* - immediately makes it much less interesting.
That depends pretty heavily on what you play the game for.
Lots of other fictional products (books, movies, etc) don't try to model a world, but just use the world as a backdrop for their real narrative point. An RPG can be like that too, and still be pretty interesting.