We did paper elections for ~100 million voters. How much more do you want to scale? Yes, it might take some time post elections to hash out results, but people don't take office the next day making that a non issue.
The benefit of transparent elections outweighs the costs of counting paper.
If we really wanted to do elections cheaply, we'd just hire a couple of undergrads to write a Node.js script and give polling stations some desktop PC's loaded up to the page. Would hardly cost anything. But doing elections cheaply is not the point.
For the USA, precinct-based paper mediated voting is the cheapest option.
We don't need PCs. Self contained scanotronic style systems (think SAT tests) work just fine. In fact, all but the largest jurisdictions (because of ballot size, complexity) could get away with manual counting.
That's cute. I'll post the same thing I post every time some non-US reader mentions how _their_ polity votes just fine with paper.
The last time I voted, I had a scantron ballot that was maybe A3 in size and needed filling both front and back. I'm probably forgetting a few offices, but I was voting for:
* President
* Federal Senator
* Federal House representative
* State Governor
* State Secretary of State
* State Senator
* State House representative
* Mayor
* City Council (slate)
* Sheriff
* State corporation commission (slate)
* Judges (slate)
* 5 or 6 ballot initiatives
* 1 or 2 local initiatives (i.e. school funding bond overrides or the like)
There's an interesting analogy to things like WMDs here. Nukes can also have a much larger impact if abused or compromised, but we still built them. I expect eventually we'll have internet voting in place regardless if the benefits outweigh the potential downsides.