OJS 3.0 is not yet at the beta stage -- we're trying hard to get the beta concretely scheduled and hope to publish a rough plan for that soon. Currently it's reasonably stable -- dare I say even usable? -- but we may well come up with some data model changes prior to the beta and won't be able to provide a user-friendly data migration strategy to cross those. There are also some bits and pieces that are distinctly unfinished: subscription management, permanent identifier management, etc; if those are crucial to you then I'd suggest staying with 2.x until at least 3.0b is released.
As for what's coming in 3.0, the biggest single thing is a comprehensive user interface rewrite. It finally looks modern. This takes the OJS interface up to the level of the current OMP release and has already gone considerably beyond that. Numerous other fundamental changes are also included: submissions are now considered to be a bundle of files rather than a single one; the workflow is much more flexible, including the ability to create roles and define how they interact with the editorial stages; etc.
The best thing to do, I suspect, is to try installing the current master branch of OJS (which will eventually be released as OJS 3.0b) to see what you can expect, both for functionality and stability, with the understanding that both are moving targets. To be clear, we will not be able to commit to supporting you if you decide to go for production with pre-release code -- but if you decide that's an acceptable level of risk, that may be OK.
As for 2.x, we expect we'll be continuing to maintain it for a considerable amount of time even after OJS 3.0 is released. This will be to smooth the transition for journals that need to fit a major change in with their own production schedules and to let OJS 3.0 mature a little bit before we mothball OJS 2.x.
Public Knowledge Project Team