Scott - thanks for the suggestion on partitioning language servers into OSGI services. I'm not involved with understanding how the language servers are being implemented too much, but I have the list of people who are working on this. In fact, this week, there is a hackathon going on with 10 people from Red Hat, IBM, Codenvy and Microsoft to work on making a headless Java language server that meets the protocol's specification. I imagine that many of the people that are working on that hackathon are not monitoring this thread. And they should probably hear about your views on how language servers could be implemented.
Would you like me to add them to this thread, attempt to set up a slack channel, or just start a new one with your private email? I think the guys you should be speaking with are Denis (IBM), Max / Gorkem (Red Hat), Evgen (Codenvy). and Sven (TypeFox).
Since the protocol is entirely controlled by Microsoft right now - any *standards* for how language servers should be implemented architecturally outside of the specification will need to be packaged as frameworks. Perhaps even net new Eclipse projects, for which I think Red Hat, IBM, and Codenvy are eager to do. While the protocol is standardized, we will need more projects that support the frameworks around creating language servers and consuming them to make this really stick.