it could be they chose this approach due to existing libraries for things like the USB comms.
I thought that too; the Linux "gadget" driver might have saved them time/cost. However, for the quantities they hope to sell in, surely it would have behooved them to just buy a license to USB device stack firmware from one of countless vendors and use a more efficient embedded processor (and a lot less flash and memory).
The jaded and pessimistic conclusion I came to was that FLIR developers on staff were more familiar with developing at a higher level. (Look at the use of Windows CE in the E series.) So, perhaps something more efficient didn't occur to them or seemed higher risk.
If the Flir ONE had a LCD, I could believe that it might be a trial effort for exploring the development hurdles in Linux versus CE (to potentially save licensing cost).
Alternatively, with interest in Raspberry PI, etc., perhaps FLIR wanted something more "hackable"? (I've heard talk from "marketing" types in corporate environments who see the "maker" crowd as guinea pigs to do product exploration on their own dime. Then, if a product shows commercial potential, they swing in for the credit.)