The note around the HA cabling would I believe refer to using the physical port on the device labelled "Failover". It should be RJ-45 on that platform, but on older generations it used a DB-9 style cable like a serial cable. In the VCMP case the two host systems should typically be configured as standalone units, and the failover port is not passed through to the VCMP guest. Think of the F5 hardware device more like a hypervisor such as ESX. If you had more than two VCMP host systems then using a switch would be recommended. It's worth noting that the IPs used in the HA direct-connection would likely not be routable from outside of the F5 cluster.
In this configuration having a crossover network cable plugged between each device allows for the systems to continue to pass data to each other for HA traffic in the event a cable, switchport, or switch fails. However, I would use the main data trunk as the primary link for any mirror traffic if your deployment calls for it. You can leverage the various failover detection mechanisms to help ensure that the right system is active depending on the failure scenario (Gateway/VLAN failsafe, for example).