Conceptual doubt

HI all,
we have been playing with bi-server in the last couple of months to test Saiku/CDE/etc with the goal to substitute our actual data analysis systems. Probably due to our background, we cannot clearly understand how the data will be managed and performance is one serious issue we would like to address.

Unsure if I will be able to explain, let's begin with our actual structure.

PostgreSQL production database -> ETL process to load MS SQL2012 fact tables and dimension tables -> MS Analysis Server OLAP cubes -> Excel users querying data

and the test structure we have been using

PostgreSQL production database -> ETL process to load MS SQL2012 fact tables and dimension tables -> Users querying Saiku/CDE/... using a Schema (defined with Schema Workbench) on bi-server

See the gap? Even if we use MDX, the data source is a set of SQL tables, not an OLAP cube. I cannot see any way to tell bi-server to use MS OLAP cubes, neither any specification on how/where/when to store and process the fact tables to build efficient pre-aggregates and the like.

In fact, we have been using exactly the same SQL2012 tables for our testing period, but these tables are not optimized for any hugue access as nobody accesses them on our actual situation. I cannot see any reason why MDX will be more efficient than SQL when accessing the same phisical structure data.

What am I missing? :)