QUESTION:
Do you know if the vSwitch/dvSwitches on other ESXi hosts forward the RARPs to the VMs connected to the given port group? The RARPs sent from one ESXi host because a VM just moved in, are they …
a) on other ESX hosts, are they seen by the VMs in the same VLAN/Port Group? (in extenso: dvSwitch fwds them to the locally connected VMs, while also updating its own MAC address table)
b) ingested/discarded by the dvSwitch, only updating its own MAC address tables?

INFO:
To other commenters, about notifiy switches not (always) working.

This is just for the case where Notify Switches is expected to help when an ESX’s uplink port is going down or especially when coming back up:

Immediately when the ESX sees “line protocol up” on the given uplink interface, it will start sending out RARP bursts through that freshly-up uplink (pun intended!).

The given switchport must go into fwd’ing state just as quickly, so be sure to turn off (in Cisco speak): Dynamic Trunking Protocol (“switchport nonegotiate”) and Spanning-Tree (“spanning-tree portfast [trunk]” | spanning-tree port type edge [trunk]” — pick the one that suits your switch hardware platform). If these are missing, or equivalents thereof, a freshly-upped switchport remains a black hole for the RARPs until it eventually goes into fwd’ing state, and that of course kills Notify Switches’s attempts to – well – notify the switches.