There are two possibilities:
1) combine them: make one session-id mechanism that caters for
   both.  Client-generated stuff in the From header seems the obvious
   choice.
2) separate them: add
 a) a server-initiated session-id mechanism to get statefull dialogs
 b) a client-generated session-id (in From) to get better statistics.
The advantage of 1) should be clear: two problems solved for the price
of one.
*Some* implementations of 2) could be better because of privacy
reasons.  a) and b) could be switched on and off independently. 
*If* browsers have a configuration screen like
   +-----------------------------------------------------------------+
    Handling of a) `statefull dialog' session-id requests:
        ( ) Always honor request
        ( ) Always honor request if it was done in a response to
            a form submission (POST).
        (*) Ask once for every site, use reply in later sessions
        ( ) Never honor request
    Generate b) statistics-enhancing session-ids:
        ( ) Yes
        (*) No
   +-----------------------------------------------------------------+
where the (*) are the default settings, *and if* a web culture
develops in which commercial sites asking for a `statefull dialog'
session-id if the browser does not send a `statistics' session-id,
purely to get better statistics, are considered rude, *then* current
levels of privacy could be mostly retained.
Some issues related to such an elaborate scheme for retaining privacy
are:
 - How do we translate the above configuration screen to something
   that can be understood by the average user?
 - Do we really want it?  This is only relevant for large numbers of
   users behind proxies accessing popular sites anyway.  Are they
   really worth the effort?
 - What happens if the makers of commercial browsers get interested in
   expanding their business to making web statistics packages, and
   start shipping browsers with default setting
    Generate b) statistics-enhancing session-ids:
        (*) Yes
        ( ) No ,
   or even hard-wire this choice into their browsers?  In the light of
   this, does it even make sense to carefully design HTTP in such a
   way that the proxy/popular_site privacy advantage can be retained?
Koen.