It is my understanding that nolock only helps when you have a write lock on a row/table. C# has something simular for accessing objects in a multithreaded environment. You can have as many 'read locks' as you want, but as soon as a 'write lock' comes in, your read locks wait until everything is commited. By putting in the nolock on a read, you by-pass the 'wait for clean data' and just read as-is.

If I'm wrong, let me know, because I thought this is how it worked.

edit: does a query have to check every row to see if it's locked or does nolock only speed up a fixed check then any waiting on writes?

eg. hypothetically, would a table that will never get written to see a scaling performance increase or a fixed; is a lock a one time check or does it have to be checked per row, every single time?