Is it just about the number of dimensions? Seems like you'd get decent performance with mongodb if each timeseries is a collection, and each entry was another document?