We might be able to use this to test different NAT topologies but it needs some work. The current NAT implementation offers a few variations but not something like a symmetric or port restricted cone NAT, which is what we're running into issues with in #33666 (moved).
Still it provides a good starting point. This ticket is going to take a lot of work.
The vnet NAT implementation appears to have the options necessary to configure the type of NAT, which they make use of in their tests. I think the next step here is to spin up two NATs and try to get them to send binding requests to each other.