Larry,
my opinion is that concurrency must be managed at both the app and the OS level -- the latter because adding additional system resources means applications will benefit from more processing capacity without requiring reconfiguration. I believe that virtualisation has a role in this, meaning that in future we will increase and decrease "available CPU" dynamically. Also, the vast pile of "old apps" must be able to benefit transparently from execution a parallel core, which requires an OS approach to concurrency.
I do believe that app programmers should have a grounding in CSP and functional languages, both of which lead to clearer thinking on parallel-capable platforms. And we need mainstream languages that do better than leaving the programmer to struggle with ThreadCreate().
I would like to see a comprehensive OS that works in a few kilobytes. I recall a Unix-like thing that boots into 3mb, it is often used in elevators and other embedded settings. Can't remember the name, but I don't think it's enterprise class (eg security?)