I think it's because the film plane, or in this case the film cylinder, wasn't vertical.
Of course, had it been vertical, then the image wouldn't have included the arches, but would instead have been a bunch of pews. Perhaps there were scenes of both, and the pews turned out to be very uninteresting. Hard to know without asking.
Here, I published a paper about the three cams that I designed and built myself,
and here, (click on the picture) you may access a galerie with a few examples of photos I took with them :
My POLKa cams have an horizontal taking angle of about 300° (4 6x18 views per 120 roll), and on the displayed picture of the rialto bridge in Venise, you can notice on both ends the same windows from a building which was actually behind me.
At the end of the paper, I show also how to watch these weird perspectives so that they look normal.