The answer to that questions depends on "the what". Some implementations are better when you have disparate data sources (flat files, spreadsheets, xml files, etc) as

DEK46656 suggests. One of the benefits of using an intermediary data marts is the enforcement of data integrity which you may lose when source your SSAS cube through database views or project data source views. This is a much larger discussion than the scope of this article.

The article itself presents the topic very well and provides scenarios in which the method can prove to be extremely useful. Kudos, Frank.