would the higher voltage cause damage or a shorter life for the first two leds? I can't seem to find much information on the maximum voltage that XM-L's can handle... does this mean that as long as I have a driver that puts out 3 amps, I can wire these in series and don't have to worry about voltage? the TaskLED driver I have been looking at has a max voltage of 24v and the Vf of a single XM-L is 3.35v...
Max voltage for an XM-L depends on how good your heat sinking is. The XM-L is spec'ed to 3A drive, which corresponds on average to about 3.35V Vf. Since all of the XM-Ls will see 2.8A, as long as your heat sinking is adequate, you should be running within spec.
The TaskLED H6CC driver is a buck driver. It will "buck down" the input voltage to the voltage that the load (the 3 series LEDs) requires. The 24V spec simply means that the max input voltage that the driver can accept safely without frying. It is not the output voltage that it sends to the load.
What you need to know if you go with a buck driver like the H6CC or the der Wichtel driver is the voltage overhead to reach regulation.
Three XM-Ls in series at 3A is about 3*3.35V = 10.05V. Your Vbatt may start at 3*4.2V = 12.6V, but quickly sag to a quasi steady state of 3*3.7V = 11.1V. At the end of the run time, your Vbatt could be around 3*3V, which is less than Vload. Thus, you won't be in regulation at the very end.
But when will the driver fall out of regulation? At best, when Vbatt falls below 10.05V (at least according to spec -- in reality, each LED has its own Vf curve, plus Vf decreases as the LED warms up). But buck drivers have a finite voltage overhead on top of Vbatt=Vload to reach regulation. If that overhead is 0.5V (which is doing good), then you need to feed the driver at least 10.5V to stay in regulation. That means Vcell = 3.5V under load. Clearly, your cells will have some capacity remaining when the driver falls out of regulation. The greater the overhead, the sooner this will occur.
You want to pick the driver with the smallest overhead (all else being equal, which is rarely the case).
Note that even out of full regulation, you should still get plenty of good light for a while. But it just won't be regulated.
Based on the analysis below, I think I would skip the 7135 approach of Download and go with a buck driver like the TaskLED or der Wichtel version.
I think I know why ma_sha encountered overheating with his build and needed to use a voltage dropping resistor.
XM-Ls have a particularly low Vf. This is a known issue with 7135-based drivers. You have to match the LED Vf fairly carefully to the Vbatt. Assuming ma_sha's XM-L Vf was the datasheet value of about 3.3V at 2.8A, then the input voltage to the driver could be quite high initially -- say 12.6V - 6.6V = 6.0V. I'm not familiar with the IMR22430 cells he used, but IMRs in general hold their voltage under load quite well. So for several minutes, the 7135 driver is probably seeing close to 6V input. Eventually, the cells will sag down to the book value of 3.7V each, or 11.1V total, and the driver might see 11.1V - 6.6V = 4.5V.
The 7135 driver efficiency is essentially Vload/Vin. So at the beginning and for most likely several minutes afterward, ma_sha's build was seeing a driver efficiency of only about 3.3V/6V = 55%. That means a lot of heat is being dissipated by the driver. Total LED power is around 9.9V * 2.8A ~28W. To generate that power, the driver draws 28W/0.55 = 51W, and thus generates about 23W of waste heat. No wonder the driver got really hot really fast.
Your 3x17670 cells should sag to the quasi steady state value of 11.1V under load fairly quickly, so hopefully you don't need a dropping resistor and won't have any driver overheating issues.
Driver efficiency at that point could be around 3.3/4.5 ~73%, which isn't that great. Waste heat from the driver could be around 10W, which is a lot. Looking at this estimated waste heat figure, you will still need a dropping resistor to keep things cool. Or you'll have to figure out a heat sinking solution for the driver.
Even with the H6CC or der Wichtel driver, I'd advise heat sinking the driver. At 90% efficiency, you are still generating about 3W of waste heat.