I think the main reason here is that it will allow you to construct your web app using the continuation support provided with Rhino. Of course, whether that's a good idea or not is debatable. This guy certainly doesn't think so
Indeed he doesn't. His reasons are:
Abandoned sessions: he claims that a flowscript in an abandoned session simply "hangs" in its last continuation, not being able to clean up the resources it holds. This can be true, but it's a bad idea to cling onto resources between requests anyway. External resources are best managed using request-scoped servlet filters and/or interceptors, so the continuations don't hold on to any external resource (i.e. a Hibernate session) between requests but have it managed for them with a session interceptor around the request. This problem is in no way limited to continuation-based state management systems.
Thread affinity: true, you lose thread affinity. This is not worse than either COM+ object model or EJB model where you never could assume logical threads of execution have 1-1 mapping to any other (VM or OS) threads. Just don't assume it.
Web farm: (aka distributing the HTTP session over several machines for failover and load balancing). Right, your script might end up running different stages of itself on different machines. But the framework must make this transparent, and Rhino-in-Spring makes lots of efforts to deliver on this promise. It uses by-name serialization stubs to mark beans from WebApplicationContext and ServletContext reachable from the continuation, and resolves them to actual beans with by-name lookup on deserialization, thus correctly rebinds the continuation to its global context no matter which physical JVM it is resurrected in. I use a similar approach in a distributed enterprise script execution system I work on for living, and the execution site transparency does wonders for scalability (we can just add another script executor JVM as we need, connecting it to our messaging middleware, and it sucks up load consisting of already running scripts immediately).
The Back Button and Branching: in my eyes, it's the best feature of all. You write a single procedural flow script, and it can handle user going back and forth and branching multiple browser tabs. His misconception is that old continuations share local variables and thus you just introduced free-range gotos into your code that operates on a single set of local variables. This is fortunately (for me) and unfortunately (for him) not true. When multiple continuations per session are allowed, then each continuation has a deep copy of its local variables, acquired by serializing/deserializing in memory, so two continuations can't interfere any way with one another. Of course, you must take care to check the validity of the non-local variables between continuations, i.e. not cache results of database queries across requests etc. But you can rest assured that your local variables are safe. They're deeply copied when execution forks.