Changed Block Tracking (CBT) impact on ESX performance?

We are discussing with a customer about the design in their VMware vSphere environment for backup.

The environment would leverage Avamar with Changed Block Tracking (CBT) for appx. 400 VMs with 20TB on a 8-pack of Cisco UCS servers.

The concern raised on CBT is that it impacts the ESX performance.

One of my colleagues found this blog from 2010:

http://virtualizationeh.ca/2010/06/04/performance-penalty-for-change-block-tracking-cbt/

I would say that the impact described there is minimal if not even not impacting anything.

However, as it's now 2 years later, can anybody share his experience in a real life environment that is similar to the above?

Or of course point me to any additional info I didn't discover by the research we did?

I would argue that any kind of backup will impact performance

- in-guest: highest impact

- VM snapshots: high

- storage / file based snapshots: impact on storage array and potentially loss of granualarity and therefore additional performance requirement on the back end even though not all VMs might need backup

- backup solution leveraging CBT: minimal and removing additional overhead on the others by reducing network load, therefore CPU cycles,...

From my perspective leveraging vSphere's CBT in such a scenario is the best, but interested to learn any other observations.

Any other points I might have missed?

Thank you!

David