What seems like years ago (because it was), some of my colleagues in Leicester started the Tagginganna project, involving inline commenting of texts. I'm not quite sure what happened to Tagginganna eventually (is it still in use?), but I do know that the concept of inline commenting - one of the "power features" of Google Docs for example - is still grossly under used.
New formats for scientific publishing such as PLOS (and latterly even Nature) have adopted online commenting as a new standard for scientific publishing. But online commenting on papers has never taken off - most papers languish without discussion in the comment section. For that reason, the launch of PeerJ Questions is particularly interesting. Although I like inline commenting, there's no particular reason to think that it would promote the frequency of comments much above burying them at the end of the paper or on another page, so PeerJ Questions uses a Quora/Digg/reddit-like voting system to try to surface the most "popular" questions.
But are questions about published manuscripts ever "popular"? In my experience most authors want the published paper to be an archival file-and-forget experience, with public discussion about as welcome as a fart in a spacesuit. Which is not to say that post-publication discussion does not go on, it simply occurs via the dark social channels it has always occupied.
You could argue that it is the high risk nature of sticking your head above the parapet in science that inhibits commenting (and is also the reason why science clings fiercely to anonymous non-transparent peer review). But if you're slightly more optimistic about the human condition you might hope to believe that comment voting will increase scientific discourse by lowering the barriers to contributions. It is no surprise to me that this innovation has come from PeerJ. If it is successful you can expect others to copy it just as they copied PLOS comments.