When you have one exit and it's blocked due to many people attempting to fit at once, the entire queue is congested.
That barrier turns one exit into two exits (even if it's backed by only one real exit). If one is congested, people can still flow through the other one, and vice versa.
This explains why the evacuation rate is roughly twice faster.
If the barrier is designed to even more virtual exits, it'll be even faster.
And there's an even faster way, that requires no architectural changes, but people will have to be trained in it so well, that they'd follow it even in panic.
And this would be some heuristic which orders people in an unambiguous way, so only one person is attempting to exit at a given moment.
I don't know, left to right order, split in rows, people holding hands... Some creative solution is required to decide what that heuristic would be.
Queues have no congestion, so they'll represent the fastest possible rate of evacuation through a single exit.
Given how badly people behave at four-way stop signs (where there's an unambiguous ordering algorithm that most drivers seem to ignore), I don't think there's much hope for a heuristic ordering working well for panicked crowds at emergency exits.
Also I wonder if a fake wall with 5 exit doors in it (close to each other), that obscured the one actual exist would speed up flow - presuming people selecting the general area of the exit first, then the detail of the exit point second. If there were 5 single doors it would separate the flow to the actual choke point, but without producing the same type of barricade.
If this research turned out to be widely applicable, I wonder if placing airline staff in front of a door would speed the exit of people through it?
I also suspect there is an optimum shape for the obstruction, and I expect that it isn't round.
All interesting stuff.
If you're interested in the subject The New Yorker has a fascinating read on the subject (CRUSH POINT - When large crowds assemble, is there a way to keep them safe? [2011]) which makes this argument. (Warning! It's long)
Hanging my head in shame for the omitted link.
Sorry for the oversight, since it's a really interesting read.