Great idea and solid arguments. I would certainly like to see this become a reality and to help out on the way should you or someone else give it a go.

However, to get anywhere near this end-goal, one would have to have many stepping stones on the way each of which must make sense to the creators and consumers at the margin (I see so many great ideas that will be awesome once there are 10m+ people on it doing whatever is the purpose of the app, but how can it still be good also when there is just 100?)

Here is what I would do. In the short term, I would not try to solve all the problems you describe nor provide all the solutions you suggest, but focus only on one: Readability.
steemSTEM is already working well to provide incentives for scientists and STEM professionals to share their work. Why not have a front end where people can post a digestible and easy-to-read summary of the paper and/or abstract, with a link to the published paper that will still require the potential reader to go beyond a pay wall. Then, I would add a new suggestion to your idea, which is to work with other STEEM communities to make it into a co-creation project. STEEM already has a bunch of great tools such as benefactor rewards to enable multiple people to contribute to posts and earn a share. With steemSTEM building a strong community, and STEEM providing a broad set of talent and the tools to let people collaborate, it could be a front end where anyone can submit a paper they would like to communicate to the public. Then, someone in the steemSTEM community who would not have to be an expert in the field, only literate enough to understand the essence, could help the person write a summary. A joining artist could help make some good looking illustrations that are both original and unique to the resulting article, and easy to understand for a normal reader.

This way, we help overcome quite a few problems without needing to take on the biggest challenge (which is publishing fully on the blockchain). steemSTEM could also aim to compile the best ~10 articles posted every month that use illustration images which we then own and text that are quality ensured by the researcher (so not dumbed down or biased too much by a journalist), and with just a little bit of formatting it could come together as a monthly publication similar to an “Illustrated science” magazine.

Anyway, this is where I see the future value proposition of STEEM. The way in which it enables frictionless collaboration and co-creation with the on-chain trust-less sharing of rewards and open data. If a significant number of scientist then saw the value in presenting parts of their work there, then who knows what one might build on top of that going forward.