I have a project I am working on at hte moment, and I am having a hard
time figuring this part out. I have for example, 10 million unique 64k
static webpages to serve. that will be served to 100,000 people (with
1 person taking a minimum of 10,000 pages). there will only be about
50% of the pages served more than once. Assuming I have enough storage
for the pages, and am using apache for the webserver. what kind of
server support would I need?. I think that the amount of threads that
can be served is based on RAM, but how much ram? etc. I have a couple
of possible ways to go with this. Either I can use blade servers, or a
rack of cheap servers. or I can make a custom set of servers using a
chip based server, as it is only serving static pages. what would be
the most efficient and cost effective. as I know serving static pages
doesn't rely so much on the processing power... or am I wrong?
I would like to know if there is any math formulae out there for this,
and what would be the options.

In your scenario, the most important factor will be the IO wait. You
need the fastest harddisks on as many controllers as you can fit in
and the fastest bus and memory you can get. Since the pages will not
need to be cached, the amount of memory is not critical, but it will
limit the number of concurent threads.

When a machine will not handle more threads, it will likely mean to
add another machine rather than more CPU/RAM to existing machines. You
might want to think about high throughput servers from Sun or IBM.