• When working in Tabular mode you need to move away from the concept of fact and dimension tables. Each of those joins will require an index creating on each table which is going to hurt the compression

    Denormalising towards a single table will give the best performance and compression

    Personally I think the use of the tabular mode should be considered very carefully. It is NOT a replacement for dimensional OLAP. You should not try and store your entire data warehouse in it.

    Instead think of it as an easy to setup analytical space for smaller datamarts which you don't want to go to the effort of modelling in the dimensional cube. You can prototype in powerpivot in Excel and then import that model into Visual Studio when you're ready to deploy