You can do what nagios does: ping more frequently, but not send an alert until several pings have failed. It guards against a packet lost to the ethers, but still picks up a failed service. Maybe ping every 30 secs, four missed pings = raise alert?