Thanks for a great article, Andy. I am in the process of designing a client/server system (.NET and SQL Server 2000/2005) in which GUIDs are used as clustered primary keys throughout the database in order to reduce the number of round trips between client and server. But then I found an article by Zach Nichter, which gave me second thoughts:

http://www.sql-server-performance.com/zn_guid_performance.asp

It seems that the two of you basically agree on the downsides of using uniqueidentifiers as clustered primary keys. But it seems that while Zach's conclusion is that this should be avoided, you are much more positive. Is this correct? And does this mean that I shouldn't redesign my system after all? I guess the answer is "it depends", but that's OK