A dimension will be the quickest way as it will continue to use aggregations across your data. If you try to limit the data against a low level data item then the cube will have to go to that low level for every query and that would affect performance

Mack