And it makes some really cool voting systems possible. For example, you can do Condorcet voting systems, which are far superior to normal first past the post. Or cool systems like single transferable vote for electing groups of representatives.
These systems are much harder to do manually and with paper ballots, but trivial to do electronically.
I'm a fan of Condorcet method and STV/IRV, but to be honest, I'd rather wait for election results that are verified over speed of implementation.
The mechanics of better election methods shouldn't require a hard-to-verify electronic voting booth.
Paper ballots are much harder to commit fraud with (and the conspiracy need to be larger) than e-vote machines all writing to memory (and likely to be closed-source).
In San Francisco, voters vote on paper ballots, and the ballots are electronically scanned. This generates a digital representation of the rankings cast on all ballots, which can be fed into tabulation software. SF actually releases these rankings publicly as a text file for each contest. See here, for example:
But even then, it's harder to do ranked paper ballots. Presumably someone has to transcribe the numbers voters wrote on the ballot into a computer, which introduces room for mistakes and lots of work.
http://californiawatch.org/files/imagecache/image-full-width...
Notice that it says (hard to read in this image) in the second and third choice that the selection must be different from the other choices.
Scotland, where I live, uses the Single Transferable Vote for local council elections. We use paper ballots. It doesn't seem to be a problem.
There was a UK-wide referendum in 2011 for adopting the Alternative Vote (single-seat STV) for House of Commons elections. It failed to pass, but if it had, that too would have used paper ballots.