if the data is completely logically isolatable, i think it'd make sense to stand up multiple clusters with multiple datastores; otherwise you'll need to have everything living in the same schema