I see that Election Canada is considering a few experiments in that direction, so here's an old piece of mine on the topic from a couple of years ago. It looks more credible as it appeared in The Mark magazine, so I've linked to that version. In it I suggest a rule of thumb: if the subject of the plebiscite isn't important enough for anyone to spend the money required to attempt a hack, then e-voting is a viable solution. Otherwise not.
Jeff Jedras makes a number of similar points, plus a few I haven't thought of, here.