Before give you some input, let me ask some addtionnal details:
1) separate data and operating system volumes of VM to different datastores (array volumes) - for different replication schedules?
You will use asynchronous replication ?
2) create one 2TB datastore and place whole VM storage in it ? - implies replication only for disaster/fast recovery of full array/lun
What is the size of your VMs ?
3) create one datastore for each VM - so be able to recover single VM from snapshot
You're talking about snapshot at storage level ?
And, your storage offer some kind of automatic data placement or tiering ?