Remember that the viewer has the worst estimation of pages being horribly inaccurate as it uses the "adobe standard" of 1024 characters per page if I remember what user_none said correctly. So being 50% of that value is certainly not unusual and far more closer to the printed edition in most cases.
The whole intent of this plugin is an at a glance estiate in your view, so you can see a 1500 vs 750 vs 300 vs 20 page type of comparison. That the 1500 might actually be 1800 and 300 is really 200 in print isn't the relevant issue. I changed the ePub algorithm so that you would hopefully get roughly consistent results between mobi and ePub formats in case your library contains a mixture of not having one or the other.
The only way to get real world printed page values is to read them off a web page like in a metadata source plugin. I would like to do this one day but it requires non trivial changes to the metadata api in calibre which are more than I can take on right now.
Btw I know this plugin will throw an error if you try to count for a drm protected book. So use quality check to exclude those.