I think a practical system is going to need a combination of approaches. At the end of the day, for safety reasons (and hack protection) cars will need to make decisions locally. There's nothing stopping them from warning each other what they're about to do, but they need to be smart enough not to rely on other vehicles doing what they claim, and recognize quickly when a given vehicle is "untrusted".
Whhile I was driving to work this morning I was thinking, what would be the biggest technological challenges to a driverless car. And in most driving situations I think a computer can navigate. However there are some situations where humans anticpate a problem may lay ahead.
For example, if I see a car drifting out of its lane 3 cars ahead I'm aware of that as a potential hazard - the computer however cannot anticipate that as a hazard?
If I see a pedestrian 1/4 mile ahead who's standing at the curb but appears as though they may dart across the road based on their body language, I slow down.
A computer cannot anticipate where hazards are whereas we can.