Amin,
The solution we usually recommend is the name as the process as a unique identifier and as id take something like processname.version. Instead of starting a process by id, you could then use a helper method that starts a process by name. This will look up the latest version of the process definition with that name, and then start the process with the id that was found. This would allow you to start processes the same way (for example when processing streams), regardless the exact version. Would this solve your use case?
This is currently not implemented as part of the code base, but if you would be interested in adding a startProcessByName method, we would definitely be interested in adding this.
Note that the approach that you tested, where you change the process definition while a process instance is already running (in this case you loaded a newer version of the process definition with the same id), is technically possible, but not recommended. The reason is that jBPM uses id as a unique identifier (so two process definitions with the same id cannot coexist) and that this ignores process instance migration (as you just update the process definition without the engine knowing, if you do this, you're responsible for making sure everything still works).
Kris