Should fact tables be in 3NF ?

Hi all,
Here is a sentence from page 236 of The Data Warehouse Lifecycle Toolkit book. "Fact tables are typically normalized to 3NF in a dimensional model because the related context is removed to dimension tables". My question is if this is a design principle or a best practice.
Suppose that I have a fact table (Card no, Customer ID, Branch ID, Date, Amount) where Customer ID and Branch ID are functionally dependent to Card no and this fact table is not in 3NF. Is this a wrong design?
And if I remove Customer ID and Branch ID from fact table and put them as attributes of Card Dimension it would lead to a snowflake schema which we all know about issues like bad performance. Any Suggestions?

Regards,
Mehdi

Take a pen and draw a line through that sentence. Ignore it. Normal forms have no meaning in a dimensional model. He just mentioned it in passing for those who worry about stuff like that. But, unless you are normalizing your model, normal forms don't matter.

I think it means that nothing should be denormalized to the fact table.