Doing this across aggregates on different nodes would distribute load across them; depending on actual workload this could be quite beneficial. Doing this across aggregates on the same node probably is not going to have any noticable effect (at least, on Linux), although in extreme cases one may hit individual device limits (mostly queue depth) both on host and storage side, so adding more device may help.
