We are seeing an identical issue with 1 host in our Cisco UCS deployment.

Two ESXi 5.1 U3 hosts

Two iSCSI vmk NICs per host

Host 2 is fine.

Host 1 is the "bad" host but everything is configured similar to Host 2.

The only difference I'm seeing is the number of paths and connected targets is different.

Host 2 we're seeing 20 targets and paths (this seems odd since we only have 4 volumes, but it's still the one that's pathing correctly so I'm not going to ask questions)

Host 1 we're seeing 16 targets and paths.

I can vmkping both NICs on host 2 from host 1.
I can vmkping one NIC on host 1 from host 2 the other does not respond.
I can vmkping both NICs on host 1 from host 1.

Things I've tried and have not worked:
I have tried to remove the vmkNIC iSCSI binding on the bad vmkNIC and re-add the binding.
I have tried a different IP address on the bad vmkNIC.

I have actually created a new vmkNIC using the same vmnic as the old vmkNIC using the same and different IP.

No matter what I do I get the same results seen in the photos at the start and below.

Again only a single host in our UCS deployment has this issue, and it's using the same UCS template from what I can tell across the board.
VLAN is correct, etc...

Good host 2, Bad host 1.

As you can see active/unused is setup correctly, this is of course flipped on the other vmk for iSCSI.

UCS traffic on the vmnics used for iSCSI A and B legs.  Obvious one isn't being used.