A June blog post by the Kindle Direct Publishing Team assured authors that the KENPC system (Kindle Edition Normalized Page Count) recorded pages read with “high precision” and that the company was constantly working to improve its “fidelity.”
What makes me skeptical here is the use of "high precision" and what there could be to improve... a page was either viewed for the time necessary to read it or not. Over all readers of a book, it should not be too hard to find out the distribution of valid read times of a page.
Unless, of course, you don't have that data, but only some snapshots of which page is opened (e.g. when a sync to the cloud servers happens).
Shame the Kindle APIs are undocumented :/