Roger,

If I understand things correctly, it isn't about equality of the Account Number match. It's about you asking the engine for more information than the nonclustered index could easily supply. The engine had a choice, do 2 searches with the nonclustered index or do 1 search with the clustered index.

Since the clustered index is physically sorted by the order of the index-specified columns, the engine knew where to find all relevant information tied to the account number. It didn't have to grab a pointer and then search all the relevant levels of the index for the additional information.

You are correct that the nonclustered index is faster when no other information (save the columns listed in the index) has been requested.