From the product documentation:
In a large network, multiple Endpoint Policy Servers can be available for an endpoint client. In such an environment, the client does an analysis from a list of Endpoint Policy Servers to find the server closest to it. The client sends a specified HTTP request to all Endpoint Policy Servers on the list. The server that replies the fastest is considered to be closest.
As to how many clients can a Policy Server support, I believe it's several thousand.
Perhaps @Oren_Shalgi can provide a more specific answer.