I didn't say it made iccp bounce (that was caused by anycast vxlan vtep configuration) which when removed the show iccp detail stabalised.
here in the case of vQFX hosts we're unable to ping their default gateway i did packet captures and the PFE/RE received the arp packets but they do not update their arp-cache for some bizzare reason.
I'm glad you sent me this link but really have Juniper not fixed this for the physical MX platforms?
I tested this on vMX 14.1 and 17.1 - no static arp entries are required so this problem is related to the vQFX per what I posted above.