You can get the best performace by doing a Retrieve with limit 1 and then a Count.
(If this is not just an esoteric question and you really have performace issues: consider using indexes!)
Are you doing anything else but counting, once you know the list is not empty? If so, you should just retrieve and check for empty. Else I would say counting is more optimized, since the query will result in a count-only and not retrieve the actual objects.
Fabians method is also a suggestion from Mendix that you can find in the Academy courses. If you haven't done the courses, I would suggest you to watch them.