For me there are two things:
- hotdeploy might be a problem (maybe only 5.x or earlier) with classloding and PermGen size, often deploy does not unload all classes
- hotdeploy looks strange and critical because if you touch one file the complete application might be restarted
- farming sounds to some administrator/customer like 'uncontrolable' because the app will restart imediately for all instances
Often a 'controlable' deplyoment is prefered in a productional environment.
If you take a node out of the cluster the farm directory is not removed (because it is not longer a cluster member ), so I suppose this is the meaning.
But if you take it out from the cluster the application should be unreacable anyway if you remove it from loadbalancing also.