Load Balanced Boost ?

If you're still using Apache, switch to nginx. In my environment, Apache can handle maximum 1000-1200 req/s (with mod_cache) but nginx can go over 13k req/s.

You will be surprised!
http://groups.drupal.org/nginx

Have a look at my server 15 hours after setting nginx (Apache still at the backend, nginx just serves css/js and boost cached pages):

No, I use mod_fcgid with suphp. But Apache is MPM Prefork as recommended by cPanel. I didn't benchmark the Worker mpm, either.

In fact, I made two optimizations at the same time: MySQL optimization with Views queries and switching to nginx on the frontend. So, indeed not only nginx made this happen.

Here is a d.o thread on the subject
http://drupal.org/node/660598

1. What is being proposed is some sort of framework that makes the database aware of which server has what cached file. If a server gets a request for a file that is cached, but it's not on this server it can copy that file locally, inject the html contents into the php buffer and close the connection. Next request will not hit php.
2. Other option is to make each write hit all servers. This is fine since for an uncached hit, the DOM is sent, connection closed and then the files are written.

Cache flushing would need to tell the other servers what files are going to be nuked. Use an md5-ed random string for authentication (like I use on the crawler).

Any other ideas on the best way this could work? I like option 2 right now.

But they still use only one server, don't they? nginx can serve boost static cache directly from memcache, too.

Technosophos has 53,900% speed up, and with a small modification, acts like nginx => memcache => boost => drupal. So the server doesn't go down again after reboot, when there is nothing in memory, thus no cache.

We've been running drupal in a distributed environment, and we have indeed found that putting nginx in front of apache to serve boost's cached files helped enormously. We are still sharing the cached files using glusterfs to have replicated cache and files directories. I've written a blog post about it here http://evolvingweb.ca/story/drupal-cloud. The gist of it is that the cache files are being replicated on all apache and nginx nodes, and adding extra nodes on the fly becomes quite easy.

On the other hand, if you are having trouble with anonymous users only, we have found that nginx typically is not the bottleneck in cpu usage, but simply in network bandwith, and indeed, it can serve 10k-16k queries per second of cached files from gluster. Adding apache nodes helps for logged in users.