That's definitely a curious effect, but the cause seems fairly straightforward. Due to the way the text lines up perfectly the algorithm seems to have misinterpreted it as two columns. By default it scans the middle third of the page for column gaps, and within that zone it seems to have found some.
If you're already using tools on your computer it's more expedient to just throw
https://www.willus.com/k2pdfopt/help/overview.shtml into the mix. The same program provides functionality for KOReader, but by using the GUI or the command line you can tailor it in much more detail and much faster. (Use max columns: 1.)