Internal virtual IP failure scenario

Using the diagram at
http://www.ultramonkey.org/3/topologies/ha-lb-eg.html for reference I
seem to have run into a failure scenario that doesn't have an elegant
solution.
 
I've set up a 4 machine cluster almost identical to the diagram and it
works wonderfully except if I pull the cable that would be connected to
the 192.168.7.2 (or 192.168.7.3) interface in the example. If this
interface loses connection there is no failover to the slave. I'm
guessing this is for lack of a ping test to the Real Server network? (I
*am* using a ping test on the External Virtual IP interface to the
gateway.)
 
I imagine the easy solution might be to add a ping test from the
Director to the Real Server network somewhere. I don't like this
solution as it depends on a single machine on the Real Server network.
Ideally, I'd like a test that actually sees the connection go away on
the Director's Internal interface and the Slave Director takes over if
it has connectivity to the Real Servers.
 
Am I missing an obvious configuration item to account for this scenario?
Right now, if I pull that cable my test cluster is down. That's a huge
single point of failure and I have seen it happen - more often due to
human failure, but still a failure.
 
Maybe pinging the broadcast and counting responses - that would be a
sloppy custom test. Even just a link test (up or down) would be nice.
I've even considered running another layer of UM3 on a couple of Real
Servers to act as the Internal ping test target but I don't want to
"anoint" any of the Real Server boxes and keep them just as dumb and
simple and toaster-like as possible so I can bounce them around, replace
them, spin up new ones at will...
 
Thanks to everyone involved for the UM3 package by the way: It's a nice
display of simple, effective elegance pulling several impressive
projects into an easy to use package, literally - Debian, in my case. ;)