Good day!
Datastore heartbeating offers a truly resiliant design. It reduces the stress on the host and reduces the number of restart attempts by identifying an active datatore heartbeat before trying to restart the VM after a management network failure.
As you seem to alude to, many multi-port NICS are actually controlled by a single chip of silicon. Even though it seems as if one is protected by redundant ports, if the single silicon chip fails, both ports will fail. You're correct in pairing NIC ports across two different chips, onboard and PCIe.
Now, with regards to the host isolation response, you have optons depending on the customer's requirements, contstraints, and expectations. The general answer here is, as always, it depends. This has changed over the years with the different versions of ESX, but today, the recommended option with vSphere 5 is "Leave powered on." This is because it eliminates the chances of a false positive and its associated VM downtime.
Imagine the result of only your management network went down had you configured the other two options. Your storage network and virtual machine network were still up. Only your management network went down, yet all your VMs are now powered off or shut down while your users could actually be accessing them. With the introduction fo vSphere 5, the default and recommended configuration has changed to "Leave powered on" to mitigate this very circumstance.
Cheers,
Mike Brown
http://VirtuallyMikeBrown.com
https://twitter.com/#!/VirtuallyMikeB
http://LinkedIn.com/in/michaelbbrown
Message was edited by: VirtuallyMikeB