When the run of TestL2LinuxBridge::test_dhcp_broadcast[#ovs-bridge] ends successfully (which means the teardown was also completed), I see that one of the node interfaces (which appeared in the common nodes available NICs at the test setup) still has
ipv4.enabled: true
ipv4dhcp.enabled: true
and an IP address (from the subnet range that was used in the test, for the VM's secondary DHCP interfaces).
This means that if you start a new test run, then this node interface will probably still not be available for use. In case you run a full network test run, my guess is that this means that we will still try to use this interface in other setups (because it did appear in the nodes-available-nics list at the start of the session), but in reality it will be unusable, which will end with test failures (which might be the reason for some of our instabilities).

More findings:
1. It happens only in the ovs-bridge scenario.
2. It happened on 2 OVN clusters I tried - one is bm03-tlv2, and the other is a PSI cluster, Both were 4.15.