Stefan - this is a pain point that many before you have encountered.
The underlying reason for this restart requirement is that the JVM running the Extension code is the same JVM that is running the Thingworx instance. Adding new code to a running jvm is a simple matter of instantiating a new classloader for the jar because you can make a some assumptions. For example, there are no other dependencies on the code (because its new) and there aren't any objects that need to be de-instantiated or rebuilt because a dependency has been updated.
Those things all need to be managed manually on an UPDATE and are certainly not trivial. The JVM simply cannot release resources that have any sort of dependency on the running environment, and as soon as an extension is loaded, dependencies get made in the JVM and the more complex the Extension, the more difficult it is to know and understand what needs to be managed upon UPDATE.
I know this doesn't help the issue, but I was hoping to give you a little more context into why things are the way they are currently and also let you know that we have spent considerable development effort attempting to remedy this for Extension developers. This is an issue that we continue to re-address and have certainly not forgotten about.
