Not that uncommon for large companies to bring their IP space in house to multihome and load-balance traffic across multiple providers. If they are routing traffic back to random data centers for centralized filtering, it makes sense how it could be coming from different regions. I'm more interested in what made your application sensitive to this. The duplicate packets does sound bizarre. Maybe a mirroring issue or firmware issue.
The changing ips were ok, it just made diagnosing the issue difficult in the logs. The duplicate requests (not at the packet level as TCP ignores duplicate packets) were the root cause of the issue.
Reminds me of the "fun" of dealing with a RADIUS system that would, about 20% of the time, send me two START packets for the same session ... with different sequence numbers.
This made successfully managing dynamic IP pool allocation on my end an ... interesting ... experience ;)
