As I said, the fact table has more foreign keys, I didn't expose them in my original post for simplicity. Indeed, if there were no other dimensions Koen's solution would be ideal as it becomes a one-to-one relationship. However, when we consider the real scenario:

Then if we apply Koen's solution it becomes like this:

In the second case, if we filter by the Groups-Accounts dimension only, everything would be correct. However, if we filter by date (20130901) then we'll have a Value of 110. This is incorrect, if you look at the original fact table the total for 20130901 should be 60.

To my understanding this is a one-to-many relationship (a single account with a single value can appear many times in the Groups-Accounts dimension), but I'm not sure if: a) this can be handled as such in SSAS without issues and performance penalties, and b) it's possible to convert this to a regular relationship to keep a star schema, which would be the ideal solution.