The behaviour of server clusters is often problematic for switched networks. What sort of cluster is this?
Many cluster technologies have two or more connections to the network but a shared IP address. Each connection has its own individual MAC address. The cluster will then share the load between the two connections by responding differently to ARP requests; different clients are told to use different MAC addresses to reach the IP address.
Normally this does not cause flooding by itself. Flooding can happen, however, when the server uses NIC A to send packets to a client, but the client is sending its packets to NIC B. This seems to be the sort of behaviour you are seeing, the client in this case being effectively your router. It is caused when the outward path is not the reverse of the inward path due to the different MAC addresses.
One way to avoid this problem, as you have suggested, is static forwarding for the server MAC addresses. But you would have to apply this at every switch.
It all depends on the exact behaviour of the cluster, and different cluster implementations behave differently. If you can find out more about its behaviour, then you can propose a solution. For example, can you say that all packets from the cluster to the router come from the same NIC? If so, you could set a static ARP in the router. However, this may affect the viability of the cluster failover, so treat it with caution.
So, to sum up: find out as much as you can aboutthe behaviour of the cluster under various conditions.