There are plenty of folks here that can give you a lot more detailed information, but just a quick note on a couple of the biggest points here. This is one of those issues that is pretty easy to understand, but it is a basic enough concept that it is hard for a new person to find a concise explanation in one particular post. There might be some "general information" threads posted around to help with these questions, but here are a couple of the bigger points.
There are different "bins" of a given LED which will vary in lumen output (along with color tint). Within a given "bin" or classification of LEDs, there will also be some unit to unit difference in output. I think Cree specs. 7% for the XM-L, for instance. But this is a fairly minor consideration, the bigger point is this (assuming we ARE talking identical LEDs, etc): Unlike incandescent bulbs, LEDs can be operated efficiently over a wide range of power inputs. The relationship between input current and output lumens isn't exactly "linear", there's usually a bit of diminishing return at the higher power levels. i.e. an XM-L at 3 amps will not have quite the output (double the output, I meant) of one at 1.5 amps. Pushing more current will require a bit more voltage, so doubling current will mean somewhat more than doubling total power used.
One of the design considerations here is that the power (or current) used as the maximum output setting may be kept to less than max. the LED is rated to handle because of the light's design. For instance, a light that uses an XM-L LED but which operates on one or two AA cells will have a maximum output setting WAY below what the XM-L is theoretically capable of providing. Trying to drive an XM-L to full output with such a power source would be too demanding on the batteries (especially if the designer wants to take into account the possibility that people might use alkaline cells).
With the 3x XM-L lights, another relevant consideration is the size of the light. If you look, for instance, at the Nitecore TM11 and TM15 and Fenix TK75. All use 4x 18650 cells, and the same XM-L LEDs (I think all cool white versions use the "U2" bin) yet the maximum lumen output increases from one to the next. The reason (I'm guess the main one, at least) has to do with the fact that the TM11 is really small, the TM15 is a bit bigger (and much more generously finned), and the Fenix is quite a bit bigger yet. 3 XM-L LEDs will produce a lot of heat when run at higher power levels. As I understand it, the TM11 and TM15 will get pretty hot pretty quickly, and the TK75 will even get pretty warm . . . but backing off a little in proportion to size keeps things from getting really hot really quickly with the smaller such lights. In addition to being hard on the LEDs if it gets too bad, the actual lumen output of the LEDs will fall off if it is run at a very high temperature.
The other factor here is the way outputs are represented. A few well-known outfits (like Surefire and Streamlight) always advertised the actual lumen outputs of their flashlights. But most of the newer Chinese companies simply reported the "calculated" or theoretical output of the LEDs. Even if this were accurate (and it might not be if the LEDs are running warm in actual use), this did not account for optical losses that you mentioned as a possibility and that, I believe, will generally be at LEAST 15 to 20%, more with inexpensive optical components. Nowadays, most of the better Chinese brands now also report actual flashlight outputs instead of theoretical LED output. If you look at the reviews, some of these reports are pretty close to reality while others may not be quite so close, but are still more realistic than the severe over-statement given by just reporting raw theoretical LED output.
But a lot of the bargain lights still report raw theoretical LED output and, in some cases, even seem to inflate those numbers (such as a 3x XM-L light sold as capable of "up to 3000 or even 4000 lumens"). From some of the reviews I've seen, it also seems pretty common that the bargain lights do not power the LEDs on the high end of things. 3x XM-L lights that can be run on one or two 18650s generally will not be pushing the LEDs at anything close to maximum (and that's a GOOD thing so far as battery life and safety are concerned). Those that use a ridiculous number of emitters - like 7 or 8 or more in a fairly standard size head - are going to produce no more than a fraction of the 6,000 lumens or more that the LEDs might be theoretically able to produce. One of the well-regarded regular reviewers here recently did some reviews on a few lights from one of the Chinese mail-order outfits, and these are good examples of the difference between claimed and real-life outputs that you encounter with lights of that type.
I've used LED flashlights for quite a while, but am still holding off on a multi XM-L monster because I've had trouble finding an "ideal" one. I really, honestly, don't "need" one for any routine purposes, so have been hesitant to drop the money on a better brand. At the same time, every bargain one I find has some weird features or idiosyncrasies that turn me off.
My advice (which should really only be taken as applying to me, not necessarily to anyone else) would be as follows. If you intend to use the light for anything important, I'd lean toward a better brand and get it from an in-country dealer that provides warranty support. Size, output, beam pattern and other important considerations can be found by looking at the reviews and posts on candidate lights. If you are looking for something because you 'want' a ridiculously bright light for the fun of it (as I currently wouldn't mind), then a bargain brand might serve just fine. There is definitely 'better' and 'worse' quality to be found in that category and one problem is that this can change from batch to batch with the bargain units. But you can find feedback - on this forum and elsewhere - that can help separate things out.
I remember muddling around with a lot of the confusing aspects of this technology, so I hope this helps . . . once you figure it out, it is not that confusing. But it does take some beating around to get to that point.
Tim