Good topic. A couple of observations:
(1) Any theoretical limit is based on a model, which makes assumptions about the thing it's modeling. The Carnot cycle puts a theoretical limit of ~37% efficiency on internal combustion engines, but that's a heat engine. There is no such limit on electric motors, which are often over 90% efficient. We've all seen the ~260 lm/W limit for phosphor-based white LEDs quoted for several years on CPF. Meanwhile, Cree is now at 303 lm/W in the lab, so they've clearly overcome at least one assumption. And they're presumably not done innovating.
(2) It might seem that once you achieve 50% efficiency, the most improvement you can get is 2x. But that's not quite true. The ultimate limiter for both flashlights and fixed lighting is the ability to dissipate heat. Suppose you have a torch or fixture than can dissipate 10W of heat. Check it out:
At 50% efficiency, max power is 20W = 10W light + 10W heat
At 80% efficiency, max power is 50W = 40W light + 10W heat
At 90% efficiency, max power is 100W = 90W light + 10W heat
At 95% efficiency, max power is 200W = 190W light + 10W heat
At 99% efficiency, max power is 1000W = 990W light + 10W heat
I'm ignoring the battery problem, of course, but the point is: going from 80% to 90% efficient isn't about the extra 10%, it's about halving the heat, which allows you to either double the input power, or run the device at the same power but generate much less heat, which means it runs cooler, which means it's even more efficient. There's plenty more goodness coming our way.