I'm not sure how implement able this would be, or the full impacts of doing something similar to what I am about to suggest - but the theory sounds good!

Currently the process of downloading, backing up, installing into a duplicate context on a duplicate database, testing and then repeating on the live system is somewhat of a long winded process. Something I see as being a big selling point would be the ability for the following:

Context 1 - containing Confluence with its own datasources ... as exists already
Context 2 - containing administration servlets and its separate datasources (in case something goes wrong with Context 1)

The administration servlets would be able to monitor Confluence, take backups from Confluence and store them separately, check for updates from the website and download them automatically (emailing the administrator when a new one is available), upgrade Confluence and if something goes wrong during upgrade be able to roll back.

Think "windows update" [download updates and tell me when they are ready to install], for Confluence (and even Jira). The benefit of this would be that minor patches could be released separately (a few times before I've been emailed an updated class as a temporary fix to a problem until a new release is done).