I don't actually expect it to be that expensive - myself and the authors of the paper have a working demo of the paper which I wrote myself in a couple of weeks (react-native clients / bigchaindb) and should be scalable to population levels (I'm still working on speeding up the crypto since I used some of my old demo code). A side note: this was my first project using react-native, and I found it to be quite fun.
As far as transparency, that's sort of the main point of the paper: if everyone can themselves check the votes by doing something as simple as looking at something like bc.info, then that imo is a large increase in transparency from the current system which only allows the room-members themselves to verify the results. I agree with you that caucuses in their current form are easy to explain, however I think we can abstract away all the difficult parts and make it a pleasant / easy experience.