GRETAP tunnels from multiple pods

I have multiple pods on a single host that each have their own GRETAP tunnels over their own eth0 to an external destination. All the GRE tunnels can be seen when captured from the calico interfaces on the host, but only 1 GRE tunnel is properly snat’d to the external destinantion. All the other tunnels are dropped. I have tried global policies that explicitly allow GRE on the workload and host endpoints, but it had no effect. Any ideas on how to get all GRE tunnels snat’d properly?