It’s possible. Sometimes it’s needed to pipeline the conversion operations, bundling rows/columns into groups, and often these groups are not contiguous, for area constrains.
I’ve seen the picture has been taken with the X-T2, which I don’t know directly (or very little), but I’m quite sure it was still a time when the rolling-shutter was not as advanced as today (namely, it was not a stacked CMS sensor, I think).
It’s also a non-bayer sensor, so the demosaicing can play a role (I’m referring, for instance, to the “halo” of noise surrounding all the fast moving parts of the picture).
…Anyhow… I’m just a bit more than guesstimating here… 
I admit I don’t know the details of that sensor.
Anyhow… you also just reminded me that I’d like to do some of this bending experiments with my X100V… It would be cool to explore that as well…