Ok, so that sounds like what I thought then (since you haven't actually confirmed this at any point) - that your book has div tags instead of p tags. As I said above, go pester user_none into seeing if he wants to make a change to his apnx code
As for switching the option, I have tried to keep this like Extract ISBN as a plugin not requiring a menu. For myself, I keep only ePub and Mobi formats in my library. I only generate a page count on the books I have "cleaned up", which means I have both formats available. I have my preference set to ePub and have no need to ever change it since my ePub calculations are "more encompassing" than the Mobi based ones as described above.
If in your case you only keep mobi files, but want the odd ePub file in case you hit this problem again, just leave the setting at ePub. All it means is that it will look at ePub formats first if the book has one, and then fallback to mobi if it does not.
There isn't really a good reason I can think of to set it to mobi. The only remote possibility I allowed for was someone who was trying to get the same numbers appearing as they would get for the apnx files generated. As you have found that isn't always a good thing currently.