Only a guess:

Are the physical switch ports to which the hosts are connected configured with e.g. spanning-tree portfast, or mode edge?

Does this happen regardless on the host to which the VM has been moved to, or can it be tied to a specific host?

"we always had to fix the issue quickly "

What do you do so far to resolve the issue? Migrate the VM back to another host,  clear ARP caches,

André