hairpin forwarding on the Nexus switch Show stopper here

Thanks for the reply. To answer your question:
1. Yes, I have about 20 VLANs, they are all tagged to the trunking port from F5 to Nexus. The Nexus switch is Layer 2.
2. I tag the native VLAN on the Nexus side, but I don't tag native in F5. Do I need to do it?
3. 
  When I ping from vCMP from Host (10.0.2.1) to Guest (10.0.2.3), by tcpdump I see the guest replies the arp:
09:51:55.095627 arp who-has 10.0.2.3 tell 10.0.2.1
09:51:55.095641 arp reply 10.0.2.3 is-at 00:11:d3:89:04:45
09:51:56.096234 arp who-has 10.0.2.3 tell 10.0.2.1
09:51:56.096247 arp reply 10.0.2.3 is-at 00:11:d3:89:04:45
09:51:57.095798 arp who-has 10.0.2.3 tell 10.0.2.1
but the host doesn't get the reply.
 Also all virtual server created in the guest cannot be accessed by the servers which connect to the Nexus switch.
 4. This is what Cisco and I assume: when the vCMP response the ping from the host, it sends the arp to the Nexus, Nexus will broadcast to the rest of 47 ports (48 ports switch), based on the Cisco (and F5) it will not retransmit the frame to the port it receives on. So there must be a way to let the Nexus ports doing so called hairpining. Because there is no virtual switch in F5 (like ESX / VMware) all the virtual servers on the guest have to use the Nexus port for internal switching. But the Nexus simply either never response or discard the arp as the flooding.
 This is the Nexus 5596, I setup the etherchannel trunk with LACP and vPC on it.  The switches seem work OK, it is layer 2 configured. 
 Thanks