First of all, thanks for the mention 
As to your questions:
* What SQL version and architecture are you using (separate database, named instances, etc.)?
- SQL 2008 Standard x64 using MS Failover Clustering
- Named Instances; separate Warehouse and Production DBs
* What architecture have you found helps in the speed department?
- We are running physical clusters that have something like 16 cores and 64 GB memory with gigabit connections to the network
- To date, I have not personally seen any performance or speed issues that would warrant us looking at improving performance. Our average setup is 6 polling engines, all at 6,500 elements or more, and we have not had an issue yet. Heck, we haven't even seen a problem with capacity that would make us want to start planning for an issue in the future.
Maybe we're just lucky, but SQL has not been a performance bottleneck for us yet.