Hi, Matt, I answered your post in the Suggestions forum as well, but I'll say a bit here as well.
It's true. The synther only uses 1.5 - 2 megapixel versions of your photos to perform the image feature extraction, matching, and scene reconstruction. For references where the Photosynth team members have said this, see this old topic for three links at the top:
( http://getsatisfaction.com/livelabs/topics/what_is_the_maximum_resolution_used_for_feature_extraction )
A simple concept for helping Photosynth recognize any particular part of a scene which I have found to be highly effective is simply to circle parts of a scene or object. This has many different benefits, but suffice it to say that the more attention that you pay to any particular surface, especially using this basic technique can yield the perfect lattice for any other images to fit into.
Please visit my "Orbital Propulsion" synth (in both viewers) for more details and a demonstration of the effects.