In my reply to your private message a couple of months ago, I explained the following:
The overall VM backup time of the entire environment will not be shortened if you simply replace 1x physical MSDP VMware backup host with 1x VM HotAdd+ClientDedup backup host, especially when the HotAdd backup host will have to go through an extra step of having to send its processed(dedup'd) data via LAN to the MSDP.
By simply replacing that with the other, you still only have one host to handle the deduplication processing.
It then follows that if you increase the number of hosts to spread out the deduplication workload, the overall backup time could be shortened.
How does one increase the number of VMware backup hosts that handle dedpulication, without having to deploy more physical hosts?
A: Why, to make use of the ESX hosts' processing resources, of course. That is if you are happy to use some of the ESXs' resources for this purpose.
How does one make an ESX host perform deduplication jobs for NetBackup during VMware backups?
A: Deploy/designate a specific VM on the ESX host as a VMware backup host, use HotAdd transport, and also enable ClientDedup.
Suppose you now have 1x physical MSDP and 3x HotAdd+ClientDedup backup hosts on separate ESXs, you now have a total of 4 hosts that share your NetBackup deduplication workload.
Suppose it used to take 6 hours for a single backup host (the physical MSDP) to backup 100 VMs, now with 4 backup hosts it could take less than 6 hours if you spread out the backup workload by assigning 25 VMs each to each of the backup hosts.
You now have 4 hosts processing the deduplication simultaneously that contribute to helping to finish backing up the 100 VMs quicker than when there was only 1 host doing everything.
The backup time won't scale linearly all the way down to 1.5 hours because of various overheads, such as the LAN transfers between the HotAdd hosts to the MSDP, the dedup ratios, the processing capacity of each of the ESX running the HotAdd VMs, and the I/O capacity of the Datastore where all the data is come from.
Regardless of whether it is SAN or HotAdd transport, the vmdk snapshots will still somehow have to be transported IN FULL from the Datastore to a host running NetBackup components.
Just because HotAdd is deployed, it does not mean that the Datastore will be magically sending out less data.