Best improvements can be seen when using low V_fwd parallel LEDs, together when battery voltage is high. For example, let's assume a medium brightness running 3 parallel SST20 LEDs, at a total of 1A current. This will have a V_fwd around 2.8V, consuming 2.8W total. At the beginning using a fully charged cell like the Samsung 35E, current draw will cause battery voltage to drop to about 4.0V. In the FW3A linear driver, the driver needs to drop 1.2V at 1A for a total of 1.2W. Efficiency in this scenario is only 70%. As battery voltage drops and V_fwd increases, overall efficiency would go up. Likewise if a different chemistry higher voltage cell is used when operating voltage is 4.1 or 4.2V, efficiency would be even worse.
In the lume driver, it will not be hard for the driver to be about 90% efficient (likely higher around 92/93%) at 1A, but that's a fair bit more efficient than the 70% of the linear driver. Not only that, power dissipated by the driver drops from 1.2W to 0.31W, saving 0.88W. This should also allow the flashlight to run cooler and marginally more efficiently (for the LED), for the same lumen output. In this case, about 28% efficiency improvement. Hope this gives a good idea.
Yes, this is true, but from what I've seen in the FW3x and FW21 flashlights, the tolerance is very tight and the flashlight is very well built; I don't see any mechanical issues and clearances are actually very generous. Again though, this concern is perhaps more related to the FW3 design; the lume topology is easily adapted to all sorts of other driver sizes and mechanical designs.