This is a bit of a shot in the dark but my guess here is that they're doing this because their stack is not able to properly deal with ICMPv6 packets on the ret

This is a bit of a shot in the dark but my guess here is that they're doing this because their stack is not able to properly deal with ICMPv6 packets on the return path. In ICMPv6 for some reason they designers saw fit to add the IP header information into the ICMP checksum so that if you're doing a NAT or other rewrite then you need to recompute the checksum for the ICMP packet, and if it's an error packet you need to do this for the inner packet as well. It seems plausible that their network stack wasn't up to the task of handling this so they sort of jury-rigged up this sort of odd connection forwarding. That's the only thing that I can think of here as otherwise there's just no planet where this makes any sense. That said, NAT for IPv6 is a generally problematic concept and they probably were flying a bit blind on how to implement it since there's no real standard way to do this. IPv6 was really designed around the idea that every endpoint would have a unique, globally routable address.