There appears to be a difference between managed bean state and component tree state. You can control managed bean state by using annotations like @RequestScoped and @SessionScoped, but it seems you don’t have a choice in whether the component tree state is saved or not (although you can choose whether it is saved on the server or client).
It seems like component tree state should only be needed for the duration of a single request as a temporary data structure to help process a request. It should be rebuilt from scratch for each request. With JSF 2.0 partial state saving makes the situation better because only form data is saved, but I don’t understand why having even form data from the previous request is useful.
If your application uses only request scope managed beans then it especially doesn’t make sense to save component tree state between requests. Even if your application has session scope managed beans I would assume that the managed beans would hold the state and the component tree still wouldn’t need to have any state between requests.
Because the component tree can be altered programmatically depending on the initial request. This is not necessarily reproduceable on the subsequent request whenever the form data has to be processed.
Further I have the impression that you think that the component tree also holds the model values. This is not true. It only holds references (by expression language) to the model values (the managed bean properties). The view state does not copy/duplicate/contain the model state. It’s just a pure UI component tree. Perhaps your confusion is based on this. Note that the term "form data" is to be interpreted as submitted values and model values.
See also: