I agree with Steve - I'm not sure you have paid full attention to the SQL in the article. He is not suggesting removing order by, but creating indexes to optimise for the clause.

Having said that, I can't remember ever creating an index to support an order by. As said, there are countless other factors far more important.

I do have one issue with the article: there is no reason the IO should have been greater for the first query than the subsequent ones. I suspect that this was actually due to fragmentation of the original clustered index that of course was not present in the indexes created specifically for the tests. As such, this measure of improvement is false.