Select Layer By Attribute will be, or should be, significantly faster since it is both compiled code as well as it pushes the execution to the data store tier closer to the data itself. That said, for 300,000 records I am not sure it makes a big difference either way.