Database Software

I'm interested in people's views on when using an Analytic Workspace is advantageous over writing the equivalent ROLAP model (i.e. tables, Materialized Views).

There are a number of high-level ROLAP/MOLAP/HOLAP comparisons posted on this wiki, http://en.wikipedia.org/wiki/MOLAP.

However, they are not specific to Oracle AW implementations.

Does anyone have a set of published guidelines on what / when AW is a good choice?
- max number of dimensions of typical small/medium/large/x-large cube
- max number of types of dimensions (level-based, recursive hierarchies, parent-child)
of typical small/medium/large/x-large cube
- max | 'right' number of dimension members in a typical small/medium/large/x-large cube
- max | 'right' number of measures in a typical small/medium/large/x-large cube
- frequency of incremental updates to the cube source
- size in MBs of a typical small/medium/large/x-large cube and dimensions
- time to full refresh a typical small/medium/large/x-large cube and dimensions
- time to incremental refresh a typical small/medium/large/x-large cube and dimensions
- runtime query of AW vs. equivalent ROLAP model?

I can only find references to Sparsity in the AW OLAP Dev. Guide, that come close to giving recommendations on the implementation (compression) of a cube, plus some aggregation level guidelines, but I can't find anything on what OLAP model / scenarios an AW might work best for, over ROLAP.

Is this even something that can be answered? given there are likely to be many 'it depends on x, y, z' responses?. Or, is it something the members of this forum spend many years learning from experience alone?

Dare I say it? what are the limitations when using AWs?

I found this White Paper:
http://www.oracle.com/technology/products/bi/olap/1450_olap10g_enhance_content_performance.pdf
"Cubes typically have 3 to 10 dimensions, but may have many more dimensions or only a single dimension." What is 'typical'?

Chris.