i'm using the standard linux software watchdog http://linux.die.net/man/8/watchdog to check against the miner logfile change-datetime-stamp (i redirected the miner output to a logfile). if there is no change for 5 min (crash) than the watchdog daemon will reboot (softreboot)

furthermore i check from the main server the status of each miner, if ssh is not responding, than i do a hard reset via my remote power switch. this remote power switch is a very high sophisticated thing ;-) it is a serial relais card with 8 relais, powering the main power switch of every minerrig - so in case the software watchdog is not rebootiing or to slow and the ssh login to the miner is timing out or rejected than the main server power off for 5 sec. and power on again. (hardreboot)

since i'm running diskless with a live image(LinuxCoin) via networkboot, the hardreboot is not a problem at all.
this setup works great for me, so i can push my setup to the limit and tweak around ... and unfortunately the server room is 10km away