The enviroment I ran AS on seperate servers (Actually 3 servers) was a terabyte enviroment. The DW had 2 billion fact records with a 3 year history. Cubes were in the 50-60 region and ranged in size any where from 250MB to 35GB in size. We had custom apps to do parallel processing of the the cubes and made extensive use of partitioning.
Without a doubt the multiple seperate AS servers resulted in a performance gain.
The decision to seperate the AS from SQL seems like it would be a simple one but it is not always needed. Many factors such as Server size, data size, number of users, cube count dim count, security and of course cost can help determine if the seperation of components is warranted.
Shoot Me! Shoot Me NOW!!!
- Daffy Duck