I think the issue of getting enough traction might be mitigated by restricting the papers that could be discussed. HN is a good example of this - very few posts engender discussions but those that seem promising (by the number of upvotes) are given a spotlight on the front page.

I think an improvement to the services you linked to would be to add a few new articles each week from very selective journals/conferences in each field. I imagine existing measures like a journal's impact score or the number of a conference's attendees would be a good start and tracking blogposts (as you're already doing) could be a good supplement.

This might help pull older or less visible publications out of obscurity; if something published in a domain-specific journal is germane to a discussion, a commenter might point this out while discussing a more highly visible article.