If I had to guess, I'd say the extra geometry is useful because, 1: the dotted pattern is less likely to be confused with user writing/drawing, and 2: they have one specific notebook that they know works seamlessly with their current algorithms without a lot of extra field testing against different notebooks.
It most likely helped a good amount in bringing this feature to market as quickly as possible, and it'll be interesting what they do with the dotted pattern concept in the future (I see it either fading into an edge case of the software that becomes a technical debt for the remainder of its lifetime as the algorithms for handling the base case of any arbitrary notebook paper improve, or evolving into the base of an Evernote certification process for notebooks).
Personally, for me this feature is the tipping point between "yeah, Evernote seems kinda cool" and "I'm actually considering using Evernote (assuming the feature finds its way to Android)". The only improvement I would make that would really seal the deal would be the ability to fax all my notes to Evernote instead (seems like much less a pain in the ass for digitising volumes of notes than meticulously photographing each page), though I suppose that as long as they support manual uploading of images through their desktop site this would be fairly trivial to script myself.
The issue is that it would be much harder for an algorithm to figure out whether a line was something you drew or the grid lines of the paper, especially if the orientation is off. If the algorithm only has to look for a dot pattern with a known relative spacing, it makes the problem much easier.