vIP gARP without Self IP in vIP range

I had always attributed that to timing and traffic patterns. Possibly the upstream device's ARP cache expired (or was very close) for those VIPs prior to failover, so that newly-initiated traffic caused the upstream device to catch the correct MAC post-failover. If a VIP was configured to maintain sessions during a failover (such as connection mirroring), 'return' traffic would prompt the upstream device to catch the new MAC as well.

In the end I'd recommend MAC masquerading, as Kai Wilke did in your post on that subject. We had some failover issues that we suspect was the upstream device not receiving or not able to process the GARPs correctly. Masquerading effectively resolved that.