According to the datasheet, a Cree XM-L LED has a forward voltage of about 3.3V at a forward current of 2.8A. 3.3V is fairly low. If the driver is 7135-based, the driver's voltage overhead to reach regulation is only about 0.1V, so in theory you should stay in regulation with Vbatt down to ~3.4V under load. In reality, this doesn't happen because the flashlight has real-world parasitic resistances (e.g., tail switch contacts, spring contacts, battery contacts), which reduce the available voltage to the driver.
When you cite your 3.9V figure in post #1, is that a Vbatt under load, or did you pull out the cell and measure its voltage without a load applied?
What brand of 18650 are you using?
Have you cleaned all of your contacts, including the tail switch? If your 3.9V measurement is an open circuit voltage measurement, then you may very well have well over 50% cell capacity remaining. If we look at hkj's 2011 discharge curves for 2A draw and assume that your cell is approx 50% discharged, we see that voltage under load could range from a high of about 3.6V and a low of about 3.4V, with the SpiderFire and AsuCell curves being outliers and even lower in voltage under load. Depending on the actual Vin to the driver at which if falls out of regulation, a Redilast 2600 might help a lot or not much at all.
I'd clean the contacts first to try to eliminate parasitic resistance, and thus voltage drop, to the extent possible. If you follow the link I provided above for vanIsleDSM's FETie switch, you'll see his curves that demonstrate the value of reducing the tail switch resistance essentially to zero.
The fact that your tail current measurement drops to 2A suggests that for your 3.9V measurement for your 18650, the actual Vin to the driver may be down about 0.05V to 0.07V under the voltage overhead required to stay in regulation (assuming that your DMM measurement didn't add too much resistance, such as from the probe wires, to skew the results). That is, instead of delivering at least ~3.4V to the driver, you may be delivering ~3.35V. As you can see, it doesn't take much to fall fairly far out of regulation.