So, I did a little playing around myself, and discovered that as far as performance goes there is zero difference. At least with the size data set that I was using.

As for ADO.NET, if you use a DataAdapter to fill a DataSet, ADO.NET won't create anything if there's no result returned. But, if you return an empty set, ADO.NET is at least able to build a DataTable. Which means, at least in theory, you won't have to check if a table was created, only that it has rows... A little less work on the .NET side.

I really like the idea that something is always returned. Even if it is just an empty set. Consistency is so nice. 😀 But really, there's not a whole lot of difference either way.