In our recently updated case study - http://aka.ms/sxp, we write about our great experiences with Windows Azure deployments and how confident we are in our ability to deploy updates quickly with no impact to our customers.  Since our services are powering parts of Microsoft.com and are receiving millions of requests per day, any downtime is highly visible.  In our case study, we write:

Availability

With Windows Azure, SXP service releases are literally push-button activities with zero planned downtime. The SXP team uses the Virtual IP (VIP) Swap feature on the Windows Azure Platform Management Portal to promote its staging environment to a production environment. Service releases have occurred approximately every six weeks since SXP launched in April 2010. The process on Windows Azure is so refined and so simple that the team no longer accounts for service releases when anticipating events that may cause downtime ...

We put this theory to the test earlier this week.  Most of our engineering and operations team were on vacation for the holidays when the recent ASP.NET security advisory was published - http://technet.microsoft.com/en-us/security/advisory/2659883  This seemed like a pretty significant issue but fortunately had a straight forward work-around that we chose to implement.  Engineering made the code changes, built the packages, and ran our test suite against the new services.

Operations was able to patch 3 services in 16 minutes without missing a single request.  This is another data point in the ease of running services on Windows Azure and one more reason we prefer to deploy on Windows Azure whenever possible.