Your reviews indicate output levels much higher than those reported on light-reviews.com for instance and even much higher than the manufacturer's specs.
My lumen estimates are based on the method
described in detail here. As you'll see, there was very good concordance with reported values from a number of manufacturers and CPF users with integrating spheres.
The key point is that the relationships are all relative, based on a calibration to available data. I make no claim as to the accuracy of all the reported lumen values that make up that data set - only that the correlation to my lightbox is highly consistent. So if my lumen estimates are "high" in these cases, than they are "high" in all cases.
Moreover, the graphs seem to indicate that the Hi (the second brightest mode) is within ~10% of the Fenix LD20 on max and yet gets nearly twice as much runtime. Can this really be, even with the improved emitter? The manufacturers themselves only claim outputs 135 lumens and 115 lumens respectively versus the 180 lumens of the LD20.
Actually, it's more like ~20% (i.e. the LD20-R4 on Turbo is ~20% brighter than the QAA-2-X on Hi). As discussed in the link above, my lightbox's relative output is not linear. If you convert to estimate lumens, you get ~180 lumens for the LD20-R4 on Turbo vs ~150 lumens for the QAA-2-x on Hi.
In any case, the difference can most likely be explained by the fact you are in essence "over-driving" the XP-G R4 in the LD20 on Turbo (i.e. pushing it past the point where efficiency begins to drop more rapidly). Emitters are not uniformly efficient across drive currents - in fact, efficiency drops off rapidly at higher currents (check out Cree literature for specs and curves). The QAA-2-X on Hi is driven a lot less hard the LD20-R4 on Turbo (or the QAA-2-X on Turbo, for that matter).
You can see this by comparing the LD20-R4 on Turbo to Hi: on Turbo, you get ~1.5 hrs for ~180 lumens - but on Hi, you get ~4 hrs for ~100 lumens. So, even though Hi is ~55% the output of Turbo, it's runtime is more than 2.5 times longer (demonstrating it is that much more efficient).
You can also see this by looking at the QAA-2-X: on Turbo, you get 3 mins at ~400 lumens, and 1hr at ~300 lumens. But on Hi, you get 150 lumens for 3.5hrs. So again, about half the output, but nearly 3.5 times the runtime.
The point to the above is simply to demonstrate the diminishing returns of efficiency as you go up in drive currents. However, at lower drive levels, the later-versions of XM-L and XP-G emitters are not very different (i.e. you tend to get similar runtime for output on the Lo-Med modes). Where the emitters differentiate is at Hi-Turbo output levels - you get either more light for equivalent runtime (i.e. comparing at most Turbo mode drive levels), or more runtime for equivalent output (i.e. most Hi modes), or some combination thereof.