It's not always a case of just throwing more CPU and hardware when intending to scale up for increased processing activities - you need to understand the components that wil be impacted prior to scaling up either horizontally or vertically.

Scheduling specific:

Primary throttle point:      

• Web Intelligence Job Server – scheduling request manager – 1 CPU can supposedly support 5 jobs – PROD currently has 2 job servers.

Secondary throttle point:  

• Web Intelligence Report Server – creates the report instances – will expand to use as many CPU’s as it requires and are available – user address memory cap of 2 GB per report server – Prod currently has 2 report servers.
  Note: Physical Address Extension (PAE) switch changes the addressing mod to enable the OS to access more than 4 GB RAM i.e. the Report Server 2 GB cap limit still applies, but you can effectively run multiple heavy processes.

In summary:

2 more CPU’s could double the current scheduling capacity but I would suggest adding more RAM and looking into PAE.
We need:

• To understand and adjust some parameters for XI 3’s built-in memory management.
  I.e. it is possible to recapture memory based on the “Enable Memory Analysis” options – I think the defaults are still in place.
• Additional info on the unmanaged disk PDF generation overheads.
• Average anticipated report run times / and sizes in case we have to adjust the -RequestTimeout parameter and the cache settings on the Report server.
  I.e.it’s not enough to just know anticipated disk space requirements as one also needs to scale parameters for anticipated load increases.
• Review Tomcat heap settings and/or Tomcat 6 upgrade.
• Review the CMS database for optimal use.   

Generic:

Web Application Server (Tomcat):

• 400 sessions
• 100 simultaneous requests per processor – non report requests i.e. folder clicks etc
• 50 simultaneous report viewing requests

CMS Database:

• 600 concurrent active users

Additional:

• If Prod is a quad / 8 GB – and you're running on a 2 CPU license, it has yet to be determined if & how the CPU license cap throttles back on a server that is scoped beyond the cap.
• Always collate detailed statistics on the existing installation prior to expanding – suggest this as priority 1.
  E.g. Use Perfmon or a similar monitoring tool over some consecutive batch and heavy user windows for OS / memory / processor / application stats.
• A review of ALL XI component configuration parameters - apart from the built in memory management mentioned previously, there are also Web Intelligence Report Server cache clean / dirty memory settings still use the defaults.
• The vendor has no recommendations for the Connection Server – possible throttle point – need to determine some thresholds on this component asap.
• Tomcat 6.0.2 supposedly has improved Memory usage optimizations / Advanced IO capabilities - the default install is 5.5.
  i.e. always look if it is possible to gain from any new releases e.g. XI 3.1 SP2 running on Tomcat 6 - benchmark this first!
• Crystal / Olap Voyager
  A total rethink will be required if intending to adopt / use these components along with scaling up for increased batch activity!
• Etc.

Most do not understand the delineations between the various offerings and it's even more disheartening to hear this from the mouths of account managers or CIO's - much to the detriment of some serious contenders for their BI space.
James Dixon's Blog takes a novel approach to rectifying this.

On the subject - Micro Strategy now offers the Report module with 100 free seats and Pentaho has some more stuff from Lucidera!!