ICEfaces execution can be divided into a number of high-level
phases. By profiling test applications and comparing the results
against stock JSF, the hotspots will be revealed, allowing us
to concentrate our efforts on the code requiring performance
optimizations. We must beware, however, as some of the hotspots
will be inherent JSF CPU sinks, giving no opportunity for
optimization.

Poor synchronization may lead to poor performance, and ICEfaces
definitely applies more synchronization than stock JSF. We
should look at each usage of synchronization to see if it is
necessary and ensure that it rarely spans user sessions.

How much CPU time is spent in each of the following phases?

 Initialization
   - does init occur exactly as necessary; i.e. we should
     make sure we're not parsing the page on every request
   JSP parsing
   Facelets parsing
 Output
   lifecycle.execute()
   lifecycle.render()
     component tree traversal
     - hot spots within components
     DOM diff
     DOM serialization
   receive-updates
   JMS update propagation
 Input
   restore view
   decode
   events