I think the mention of JVM start up time is a bit erroneous. A mechanism for uninterrupted service during deployments is needed whether the JVM takes a millisecond, a second or a minute to load up.

The article is interesting though.