I am a big fan of nolock and have used it very successfully (and prudently) over the years. I must admit that this whole thread has made me reconsider how I use it.

There is one thing that has not been clarified. In the case of duplicate rows (which causes me the most concern and is something that we have seen a few times in our environment), does this behavior only occur when 1) a clustered index is updated and 2) a non-sequential clustered index is inserted against? I noticed that Tony Rogerson's example of this involved updating a clustered index, causing the row(s) to physically move because of a page split. Could this behavior occur with non-clustered index modifications as well?