Deprecated - Performance and scalability

My company is considering migrating it's website (currently static HTML) into Drupal 4.7, and we have a few concerns.

The scenerio: We record, process, and provide data for the seismological community. On a day to day basis, we receive a fair number of hits from both the general public and researchers accessing our data. Yesterday, for example, we had about 15,000 visits, 250,000 hits, and used about 5 GB of bandwidth.

For those looking to have Drupal serve private files efficiently, I've posted a simple core patch that adds X-Sendfile support:
http://drupal.org/node/74472

Hi,

I am interested in knowing how well Drupal would scale in the following examples (taking into account a number of custom made but well coded extra modules).

a) User contributed global news website, a *bit* like a mash up of Wikinews / Scoop.co.nz.

b) User contributed global community social networking site, a *bit* like a mash up of Facebook / Craigslist / Google Maps

In both examples, assume a very large amount of traffic, like ebay.com or wikipedia.org.

Could anyone used to patching patch the latest i18n.module (http://drupal.org/node/5917/release, 4.7.0. or CVS version) with the patch posted here: http://drupal.org/node/65801#comment-105323, and post the patch in that same thread, so also others with the same problems can use it.

As requested in the forums http://drupal.org/node/73159#comment-136559 by rapture:

"Looks like we just survived a front page digg. Drupal.org admins want to show any stats?"

And if you read the "Case Studies" section of the performance book:
http://drupal.org/node/44740

The Case Studies leave a little to be desired - so, what kinds of things seem important to gather about this?

Recently having read Dries Buytaert's article on Drupal's performance relating to the perfomrance of different Drupal/Server configurations (http://buytaert.net/drupal-webserver-configurations-compared), it raised a question, in my mind, regarding how many servers would be required to service a large Drupal implementation which was mostly based on authenticated request or at least a large number.

In order to try and get some insight into this I sent an email and posted a question with ourmedia.org who I believe, currently have one of the largest Drupal-esque implementation as I was interested in their experiences. Unfortunately, I have not had any feedback from them as of yet, so I thought perhaps this forum could be of some assistance.

The article above is based on some basic benchmarking, however it does raise a question of how effective Drupal would be in an implementation which was mostly based on a very large number of authenticated requests per second. The benchmarking shows that at best, in the configurations and hardware used, Drupal could handle 3 to 6 requests per second, different hardware would obviously have varying results, but as a baseline it is a good indicator. Does anyone have any experience with an implementation that has been mostly based on authenticated requests and has had to handle large volumes of traffic?

We currently use Drupal on a couple of small projects and I have already seen the effect the load multiple requests can have on the CPU usage. We have 3 PIII webservers hardware hash load balanced and one a few occassions, we have seen some CPU flatlines, hence I found Dries' article very interesting. Although I am a great fan of Drupal, it is sad for me that it may not be fit for purpose for a new project that we are currently planning. This is not only because we will have to either look at some other solution or go for a ground up approach, but also because Drupal is a great product and the of the flexibility it provides.