I agree with the microsoft approach... but mainly because when I design tables... I reference them through the Identity keys from the get go. Therefore having a clustered index on a sequential ID provides a double benefit when you actually reference a table through the ID key. For name or address searches, I recommend non-clustered index keys that taylor to both the where and select clauses of a common SQL search. The main thing you want to avoid is bookmark-lookups, because they are expensive.

But hey, if there was only one right way of doing things then we wouldn't need programmers or DBAs.