If you have a circumstance where 2 lanes are merging into 1 lane, merge as late as possible. It sounds counterproductive, but it's actually best for traffic as a whole.
See http://trafficwaves.org/seatraf.html and the HN discussion at http://news.ycombinator.com/item?id=785259 . http://drmomentum.com/aces/archives/003800.html also has a number of nice links.
In gbadman's example one lane (let's say the right hand lane) goes to a road that does not back up, whereas the left hand lane goes to a road that will back up through the interchange. In this situation, you have a left lane congested with people waiting to go on the busy road, and a right lane that should be almost free flowing. By most understandings (including the execellent book, Traffic) people wanting to go to the busy road should wait in line in the left lane, allowing people going to the less busy road to pass freely. However, many drivers will stay in the right lane until the very last minute, then attempt to move into the left lane before the right lane splits off, backing up both lanes.
I think the lane split is the better example of the prisoner's dilemma. As others have mentioned that in the two-lane merge situation the global optimum is the selfish approach.
I think it's important to stress the caveats of that recommendation.
It applies to heavy, congested (as opposed to flowing or merely partially congested) traffic with a merge that is orderly, zipper-like.
So, don't just merge as late as possible. Rather, when faced with heavy, congested traffic perform an orderly merge, as late as possible.
As i mentioned in some other comment, self-driving cars taht you "hail" are the best compromise between public transport and private transport.