Oracle has posted a very nice white paper on Oracle Technet that shows how Oracle OLAP on Exadata can be used to support adhoc OLAP-style queries on a half-rack Exadata box. The white paper builds on the SalesTrack analytic workspace with 2.2 million customer dimension values, and 1,000,000,000 fact rows. Fifty simultaneous queries ran against the database, simulating many more business users--no "think time" between queries.
The most interesting part for me is the appendix that shows the scripts used to build the queries and the settings used to put the analytic workspace into Exadata flash cache. This becomes the Oracle standard of how to implement cubes on an Exadata box.