Both makes sense.
Just a few thoughts from my recovery biased point of view.
For a 5TB vmdk you need a datastore that is even larger than those 5 TB.
I often see cases where a datastore needs to be rebuild from scratch - means copy everything out - reformat - and populate it again. With datastores of 5 TB or larger this often becomes impossible as the time is too short to do so in the weekend maintenance window. In practice this occasionally means that maintenance tasks are not possible as production VMs that also use the same datastore have to work on monday morning.
This would not be a problem at all if you create a NTFS-LUN instead.
So I would say the 5TB vmdk is more convenient to handle, the NTFS-LUN via iSCSI is safer.
If you go the 5 TB vmdk way - create the vmdk as eager-zeroed and run
vmkfstools -p 0 name-of-5tb-disk-flat.vmdk > 5tb-mappingtable.txt
Keep that file - it can make the difference between desaster and "we lost some nerves but could avoid the worst"
In the same context I always recommend to size your VMFS-LUNs so that every single one can be evacuated during next night.
