If you are working with a VMFS datastore, one of the typical questions is how many LUNs will I need? Considering that each storage LUN will be formatted to become a VMFS datastore and that there is a 64 TB max limit for datastore size, potentially you may use only one LUN.

But in vSphere HA, datastore heartbeats require at least two datastores, although you can configure some vSphere HA advanced settings to use just one, or just don't use it (depending on your storage specifications).

Another aspect is related to a SCSI reservations issue for VMFS datastores; the larger your datastore, the more VMs and VMDKs and more SCSI reservations you will have. Limit the number of VMs (there isn't a magic number, but in most cases it is around 30–40 VMs) and make the number of required datastores bigger. That means more time for re-scanning and more complexity to manage them (for the second aspect, SDRS could be a good solution).

Finally, you have also to consider the different paths that you can have and remember that one per time could be used, also with a round-robin that simply distributed the requests, but at a specific time, one path is used. With more datastores, you can better distribute the active paths and increase the fabric utilization.