Basically, you don't want to lose any of your detailed data, to allow for the drill-down OLAP operation.
In a data warehouse, the grain of say, customer orders, would be order line item
, ie the most detailed.
What you should do is to figure out which aggregates to pre-calculate, and use a tool to do automate that for you. The aggregated data would go in its own tables.
A smart OLAP cube will realize when you should use an aggregate and re-write your query to use the aggregated data instead.
Check out Pentaho Aggregation Designer, as well as Mondrian OLAP server/Saiku pivot tables. All FOSS.