>I think the fact that it's a distributed model is the scariest part! One of the biggest challenges of creating the new apps will be exception code for those distributed services that aren't working.
>

You handle these as you would any other complex system- each subsystem is responsible for reporting errors it encounters while attempting to acquire resources. If needed the component passes the error information back to the component requesting its services, and on down the line. The chain of responsibility pattern makes this seemingly huge task totally manageable.