That would work if this dimension was the only one present in the cube. Imagine I duplicate the values for Account 2 and 3 in the fact table, each with its own unique key to the dimension, and create a regular relationship between them. This is easy and works beautifully, but if I add the time dimension when filtering by date the sum would be more than the real sum because there are more fact records than before. I could tell the client something like "hey, you have this account 3 times in the dimension, so it's normal that when filtering by date you get values as if the account was triplicated", but I don't think this is a viable solution. Another one would be doing two cubes, one with only this dimension and one with the rest, but this is ugly as well.

I can't believe something like this can't be properly or directly addressed by SSAS. Dimensions with duplicates are very common in all kinds of business. I understand the duplicate key issue, but I don't understand how SSAS doesn't do something internally to handle this and consider each duplicate as unique. I've been trying the many-to-many relationship and the damn thing doesn't support unary operators (another requirement) properly, so this solution is out too.