Well, you can do an indicational test. Take a look at the post info from the ISI (rfactor2) engine (which is different by the way then the GSC engine).
96 Kbits/sec per user for server populations less than 10
256 Kbits/sec per user for server populations between 10 and 20
768 Kbits/sec per user for server populations between 20 and 30
So, a bit simply said,
10 users, multiply the usage of 1 driver times 10.
20 users, multiply the usage of 1 driver times 15
30 users, multiply the usage of 1 driver times 25
Reason for this increase, is the fact that positional data is more with more drivers. (note that its not increasing evenly)
What I did in the beginning of my server, was doing test on a public server.
Install a piece of software that monitors the usage of your network card. Turn off all other applications where possible, and take a loot at the usage over 10 minutes. This is your default data (so outside your dedicted server), which is being run. Think in checks to windows update, antivirus polling, stuff like that.
Now set up a public server, with an interesting car/track combo. keep it running for a while, and chech on your network usage.
With a little math, you can make reasonable accurate calculations about the use of the dediced server. Divide it through the number of drivers, and you have an indication of usage per user.
To be safe, I advise you to take at LEAST double capacity. So if you calculations tell you you need a 2Mbit connection, take at least 4. This will cover the dips in speed.
Next up, running extensive testing, so over a longer period of time. Join your own server, and ask people if all is going well.
That way, you should end up with at least some kind of data.