I think that in this particular case, Fuji put a HUGE amount of effort into the JPEG engine for the X-Trans sensor. De-mosaicing that pseudorandom array requires new algorithms and I suspect a LOT of processor horsepower.
Under most conditions, in fact, NONE of the stand-alone processors (talking to you, Adobe) does as well as consistently as the in-camera engine, given default presets. What this says is that what Fuji is doing in that RAW engine is DIFFICULT, even for experts, and all the more so in a compact battery-powered computer (even given a lot of VLSIC* optimization). This camera is not like cameras with Beyer arrays, the demosaicing of which involves generallty mature, well-understood, and highly optimized algorithms.