Question : How many servers would I need for 100 concurrent users for terminal services?

I have a company that is looking to have all their users connect to a terminal server at a data center. I am just trying to see if I could do this with one beefed up server (Dual quad core xeon, 16gb RAM, etc) with say a 5MB pipe going to it? Or would I have to do 2 or 3 servers and load balance them? The users are only going to be using one Application so I would follow the standard 64-128MB per user. Also I know with Server 2008 Standard, is that the way I wanna go?

Answer : How many servers would I need for 100 concurrent users for terminal services?

While the configuration you describe could probably handle the type of use/traffic you are describing - there are logistical/operational drawbacks to such an environment.

A drawback of 32Bit MS Windows OS's - regardless of whether it is Server 2003/8 - a 2GB limit applies to the kernel - meaning each and every TS user will be sharing the same 2GB kernel - realistically this affect how many TS sessions can be "usefully" active - which in turn is determined by the "client" application's reliance upon calls to the kernel.

By far and above a properly load-balanced mulitple server configuration will provide a better user experience not too mention a greater degree of fault tolerence - which will help you as the provider in the longer term.

2 physical servers (regardless of processors/cores)  are better equipped to handle 100 concurrent users than one physical "mutliple" processor based box.  The contributory factors to efficiency go deeper than just CPU/RAM & connectivity - you must take user and application response/reaction time in to consideration as well, that in turn, will load the connections and available resources.  

As a final thought - I am assuming that this is likely to be a  mission critical environment - and therefore - if a single server goes down the entire company/client connections go down with it - whereas a mulitple "physical" server setup would at least provide continuity while a faulty server is repaired.

Splitting 100 active and concurrent users across 2 or 3 physical servers (be they dual/quadcore) will without any doubt give a more reliable and efficient user experience than a single server with either 1 or 2 Quad cores.