One idea that i have thought of in an RPG to deal with combat and player killing is having a police force of kind, an organisation that you join that has an automated list of player killers, and worst offenders and then the in game PCs go and track them down and killing them, or penalising them. It would then allow for player killing, but add a reason not to.
As you are trying to have player killing though it might be a better idea to keep it within a city, ie of you kill a player from your city then the city guards come after you, but if you kill someone from another city then you might be rewarded, unless the two cities are at peace and so on...
I think you get the idea.

I guess to answer that I''d have to know what the risk/rewards would be for either scenario.

City-scale conflict is great, but becomes quickly mind-numbing and pointless unless there is a factional reason to join combat and ensure your side is victorious. Bonus modifiers for other aspects of the game and access to shared areas that have their own set of rewards are just two methods already used that I personally think have worked well. It draws people into the conflict knowing that failure to defend can have global consequences, while a successful attack benefits the entire faction in both long-term and short-term rewards.

Roving groups, or bandits, or just gangs, are purely instant-reward scenarios. These are usually the groups who like the hunt, the feeling of power that comes from the kill, the personalized aspect of small-group combat. Looting is usually the benefit of this type of PvP combat - again an instant reward.

Ask yourself this: which fits the setting of your game more? Are you trying to balance PvP gamers with those who don''t want part of the conflict, or is everyone open to combat - depending on where they are? Does winning a city-wide conflict offer protective areas, or is it simply better to band together in smaller groups while exploring/roaming?

I get what you''re saying about preferring to halfway implement something the public wants over fully implementing something no one wants, but I would think it easier for people to express what they''d prefer based more on your game mechanics and design.

My 2 bytes.