It's all in the protocol and implementations: you don't need to trust huge TCB's and manufacturing of specific device. Low TCB, diverse, clients doing the checking is all you need. The paper comparison is unfair given online and electronic is assumed in the requirements of the OP. Something closer to paper in usability, cost, or security would be:
Scantegrity, like other secure electronic systems, scores low on usability[1].
Recently the CEO of Fox-IT reported a delay to our minister in the Netherlands[2] with researching the specs for a usable and secure vote printer and vote counter[3]. It's not as easy as people think, especially the elder seem to have trouble with it.
Thanks for the data. I'll factor that into future comments. The direction that leads, though, is improve the usability rather than toss out the whole system or its principles.
My main recommendation if anyone asks with intent to deploy is paper and optical readers. Cheap, easy to use, easy to check at booth, and easy to audit later. I prefer computers stay out of voting as much as possible. However, if they're there, Scantegrity line and Civitas seem like top contenders to build on.
Also note in any analysis that secure voting has so many seemingly-contradictory requirements and attack points that any solution will likely pose difficulties. I expect some responsibility and effort from the system's users just like they must put effort into learning to drive. That said, the usability can certainly improve and we should put every effort into that.
http://www.scantegrity.org/