I had the same problem in my nested lab and it turned out that my MTU size was configured correctly, but in between a switch could not handle jumbo frames. Have you pinged all TEP IP addresses from on of your ESX server? ping ++netstack=vxlan <dst IP> -s 1600 -d
-s is for size, -d is for don't fragment
How is your LAB structured? Are the edge VMs inside your nested lab or outside? If the EdgeVM is inside, you can get problems, because Host TEP and Edge TEP are in the same VLAN and this is only supported under certain conditions with one Network Adapter.
