Wait for JC to state that "We don't use IDENTITY, Cobol-like monster from 50's" :hehe:

The only place where I would agree on legitimate use of GUID's for PK, would be systems where PK must be generated by application before inserting data to database for one or another reasons.

In the rest - I would use surrogate IDENTITY PK's.

Decision on clustering is a bit separate, but mostly you would want this one clustered as well.

With new Oracle-like SEQUENCEs in SQL2012, use of GUID's is even less prominent, as you will be able to get PK values from SEQUENCE independent of inserting process.