Why do you need to deploy 3 VM for APIC-EM redundancy. If I have 2 ESX Servers, 2 VM are enough
This is based on a fundamental concept in distributed computing called "split brain".
If one VM is not able to see the other, it could be due to the other node failing or the link to the other node failing.
In the scenario of the link failing, both sides could potentially continue to process requests (split brain) which would cause issues when they were reconnected. Which version of the "truth" would you trust. To address this, if a cluster is partitioned, it needs to have >50% of the members active.
The minimum cluster size you need to do this is 3.
Which is why 3 is usually the minimum number of members for any HA cluster.
It is also why clusters often have odd numbers of members.
With APIC-EM, if you only have 2 VM in a cluster, and one VM goes down, the cluster goes down.