JSF 2.0 will add the ability for applications to visit all or a subset of the components in the component Tree in context using a visitTree function in a similar manner to how invokeOnComponent works with the following improvements:
1) For a single component, visitTree is faster than invokeOnComponent because visitTree avoids most calls to getClientId() on the traversed components
2) For multiple components, visitTree can visit all of the components in a single pass, avoiding replicating the work to setup and teardown the component contexts when visiting. In addition, visitTree guarantees the relative ordering of the component visits
3) visitTree supports optionally skipping non-rendered components, matching the visiting behavior of the JSF phases.
4) visitTree supports components and renderers that set up different or additional context when visiting during the encoding phase
With a few improvements, this visitation scheme can be extended to support optimizing partial page rendering. Trinidad currently implements partial apge rendering by performing a full render of the component tree (with optimizations for NamingContainers with no targets) and discarding the results of the content outside of the partial targets. This is inefficient because the unrendered components still generate their markup, more expensively, they continue to evaluate the EL and execute the models necessary to generate that content. Thus, on a large complicated page, it is not especially faster (on the server) to render a small portion of the page than a large portion (with the exception of the NamingContainer optimization which does typically avoid executing tables). This limitation is one of the main performance issues with PPR (the other is the need to re-execute the JSP in JSP environments)