Does anyone have some tests that show that there will be a performance gain from the following example:

"In a Custoemr table you have moved the City and State fields to a related ZipCode table with the join based on the ZipCode in teh customer table. You did this to achieve 3rd Normal Form (All fields are dependent on ONLY the PK) when you realized that City and State were codependent on ZipCode and the PK. Your system is used to print 10,000,000 mailing labels once a week. Your tesint finds that a lable printed from a two table join as comapred to a single table operation takes an exteremly longer time to process. You determine that teh label printing is more of an isue than the update anomaly possibilities so you decide to denormalize and put the city and state fields in the customer table directly." -- Jim Booth

I tried to demonstrate this to someone, and was left with the conclusion that it is slower to put the City and State in the Customer table. It makes sense to me, because in a 10,000,000 record Customer table, there will be significant duplication of City and State, so keeping it in 3rd normal form results in less disk i/o, assuming that you are using a database (like VFP) and OS that cache indexs and data in ram.