The main obstacle to dethroning commercial journals, as I see it, is the reputation they've built up over the decades. They're simply prestigious by default because of lock-in.

There are now ways of running an online reputational community, e.g. StackExchange, but they rely on a willingness to do things a different way. The more conservative members of the academic community aren't likely to shift their approaches, especially when they don't already participate in forums or academic blogging. They don't directly feel the pain of expensive subscriptions, either. So the academic publishers - the old order - can continue to rely on their deeply entrenched position as the arbiters of research quality - even though this quality is created by the unpaid academics who sit on the editorial committees.

(In fact, the publishers even sell this as a value that their journals provide - reputation - when it isn't an intrinsic value, but one that accrued to them while there was no feasible alternative. That value has now lost its underlying basis, but it persists because of inertia, and they're merely extracting rent from their ownership of the journals' brands, now.)

The solution, to my mind, has to be multifold:

1) Spread the dangerous idea that science should be accessible to all and free from paywalls, such that in all polite circles it should be a matter of course to believe so;

2) Build a very simple-to-use, cross-journal meta-database and community - preferably open-source - that collects papers from all open journals and aggregates upvotes, citations and prestige/rankings, and help it gain wide, career-influencing acceptance like Github enjoys now (maybe the arXiv could be evolved and adapted to fill this role?);

3) Get buy-in from senior decisionmakers and thought leaders in various academic fields by offering them a chance to shape the system.

This won't happen without the joint efforts of leading academics (like Sir Timothy Gowers), skilled technologists, UX designers, and government funders. Most of us on HN know how we could build a technical solution that fills these needs - we might even volunteer our time toward it - but we need marketers and advocates who can sell this system to the people who matter (who currently don't see the need to upset the current system). The reputational advantage enjoyed by commercial publishers won't be overcome simply with engineering.

(If anyone knows of open-source projects to this end, or about the arXiv's plans, please share them.)