It seems that the same group ID 0 is used for different flows - I believe it is being overwritten every time. In the example below group 0 is setting the MAC for 10.0.123.4

cookie=0x8000003, duration=5464.162s, table=21, n_packets=930, n_bytes=91140, priority=42,ip,metadata=0x222e0/0xfffffffe,nw_dst=10.0.123.2 actions=write_actions(group:0)
cookie=0x8000003, duration=4572.631s, table=21, n_packets=930, n_bytes=91140, priority=42,ip,metadata=0x222e0/0xfffffffe,nw_dst=10.0.123.4 actions=write_actions(group:0)
group_id=0,type=all,bucket=actions=set_field:fa:16:3e:53:c4:2b->eth_dst,load:0x700->NXM_NX_REG6[],resubmit(,220)

Looking at the log, there is an error each time a group ID is trying to be allocated by the Genius IdManager.
It seems that besides this regression causing connectivity issues, the error handling here is wrong - the group ID 0 should not be used, but rather a clear error should be outputted and no group should be configured.

2016-08-16 11:45:17,927 | ERROR | n-dispatcher-576 | IdManager | 313 - org.opendaylight.genius.idmanager-impl - 0.1.0.SNAPSHOT | Allocate id in pool nextHopPointerPoo
l failed due to java.util.NoSuchElementException: Specified pool Optional.absent() does not exist