Description of problem:

When Jumbo Frames are set for infra, POD to POD in virtual OCP cluster has degraded network performance.

Version-Release number of selected component (if applicable):
4.16+

How reproducible:
Multiple cases able to reproduce

Steps to Reproduce:
Create baremetal cluster ocpv cluster with ocp cluster running in vms.

9000 - OCP-V bare-metal hosts
8900 - OCP-V cluster pod network
8900 - VM interface
8800 - virtual cluster pod network

Actual results: Using k8s-netperf

• Pod to Pod 52.535384 (Mb/s) // 8192 message size 45.809499 (Mb/s)
• VM to VM 2414.267904 (Mb/s) // 8192 message size 8152.709632 (Mb/s)

Expected results: Similar speeds for pod to pod

Additional info:

Degraded network performance between pods on different nodes on different OCP-Virt hypervisors running on same virtual cluster pod network. Seen when using jumbo frames.

```
9000 - OCP-V bare-metal hosts OCP-V bare-metal hosts

v
8900 - OCP-V cluster pod network |

v v
8800 - virtual cluster pod <--- Degraded Perf---> pod
```

Disabling offloading for UDP checksums has no effect.
Also other offloading disabled or enabled has no effect