The first issue seems to arise as (for reasons that are far too boring to go into) the active interface on all the machines is eth1 (lo is also active but eth0 is not). In this case starting condor results in all the daemons binding to the loopback address rather than eth1. Adding NETWORK_INTERFACE = xx.xx.xx.xx to the config file acts as a workaround and the daemons bind to the correct interface.