The idea of distributed personalization is to leverage the browsers' capability to compose a page from different sources using either the script tage, frames, or iframes.

The concept is simple - since the peice of the page can be brought from any where, bring it in from a machine dedicated to that peice of functionality.

For example, you could set up a weather server.  All that it does is manage weather data.  It has a web app that gives and reads cookies, or takes the arguments in the form of cgi query string.  The app returns the appropriate html and javascript to render the personalized data.

The query string/cookie could just be the user's id.  In which case it is up to the app server to remember that user's preferences.  In this manner the main server serves simple templates.

Or the main server could get the users' preferences itself from a central repository and pass the descrete preferences to the app servers.

In other words www.com could look up glenn's prefs, and display a template with 

	'script src="weatherserver.com/forecast?user=glenn"'  

and 

	'script src="stockserver.com/quotes?user=glenn"' 

which is lightweight for the main server,  or it could look up the preferences and pass those to the appservers like so:

	'script src="weatherserver.com/forecast?zip=37931"'

and

	'script src="stockserver.com/quotes?stocks=SSP,QQQ,VIGN"'

Since we're talking about plain old web methods, existing load balancing and clustering solutions can be employed to scale weatherserver and stockserver.

Further, the mechanism for selecting the personalized data could be via cookies, which would allow static sites to take advantage of the personalization system.