Odds are that this is because an online pdftohtml utility was used, I've seen a few that create more accurate output than what Calibre's current pdftohtml can do, but they use a lot of hardcoded divs like these to maintain the original layouts more precisely (unfortunately simultaneously killing reflow)
Some examples:
http://convertpdftohtml.net/
http://www.pdfonline.com/convert-pdf...l/default.aspx