hopefully html5 pushstate will alleviate a lot of the complications with providing full fallbacks
As it sits, we use a full client side UI library and push all UI concerns into the UI, they communicate with the server via REST to get data and the client is responsible for rendering that data. We research our numbers extensively and the cost analysis of changing our development model to support clients that cannot support full JavaScript clients supports the conclusion "at least for us" that a progressive enhancement development model is not cost effective.
The money recouped with the shortened development cycle of Ajax clients outweighs the amount we would receive chasing some fraction of < 2% of potential clients, If we get to the point that we need to chase that 2% for growth we are a) doing great and b) would be better served segmenting that traffic to a completely independent server framework based UI that it built for those clients.
It is my belief that mixing the two development models gives you the worst of both worlds when it comes to features and maintainability.
Full (unbroken) link: http://calendar.perfplanet.com/2010/thoughts-on-performance/
Both of these can be achieved using progressive enhancement. There is no technical reason to include it in the page itself.