Simply put, they are all wrong. A bottleneck is a bottleneck and there is no fairy dust magic queueing technique that works better than another. At best, some techniques might have a behavioral advantage...but I stress again, they are all inconsequential. We are talking tens of seconds of time savings at best.
Ignoring early exits, at steady state, the only thing that really matters is flow _after_ the merge - that's what determines the rate at which traffic passes the restriction.
However, it's a simple matter of utilization, if everyone waited properly in line, traffic would seize up in Beijing as many intersections would become blocked by single lines of cars next to empty unused lanes.
We are talking about saving minutes, 10s of minutes, or even days [1]. Funny how Americans haven't seen real traffic before and think they know the right global solutions.
[1] http://www.npr.org/blogs/thetwo-way/2010/08/23/129376194/tra...