Thanks for the reply, Francis.
Let detail more the scenario, so you understand what I try to do.

[ 1.localserverip (main load balancer, with nginx and php-fpm installed) ]
|
---- + [ 2.localserverip (node with php-fpm installed) ]
|
---- + [ 3.localserverip (node with php-fpm installed) ]
|
---- + [ 4.localserverip (node with php-fpm installed) ]

A better graphical example can be viewed in this image:
http://farm6.static.flickr.com/5023/5631528237_9153f0d4a5_o.gif

Basically, 1.localserverip is the site entrance, where nginx load balancing is done.
Each server (including the 1.localserverip) will have the exact same content into /var/www/html directory.
The served content consist in static (images, cass, js, etc.) and php files.
So if an user access the site, 1.localserverip will pick a node and serve the content, based on the load balancing setup:
upstream proxy {
server 1.localserverip:8000;
server 2.localserverip:8000;
server 3.localserverip:8000;
server 4.localserverip:8000;
}

My goal is to be able to load balance not only the proxy but also the fastcgi.
What would be the configuration like?