You need multiple cameras seeing every image point so your CV algorithms can have a good depth reconstruction -- and so it doesn't break if one camera has a lens flare. >2 viewpoints also helps with reconstruction ambiguities, etc.

Google Jump uses ~16 cameras for the same reason.

I would imagine the stitching required a good bit of overlap of a video with its neighboring videos.

based upon the photo-spheres i've made, the more pictures definitely help correct light-levels. Especially correcting areas looking at sun or glare.