For us, this is about high availability rather than load balancing. I think we can handle this in our application layer by switching to whichever tracker is online, and some script on each tracker machine to watch for a tracker to be down and start up if appropriate.
Should be fine as long as the other tracker is completely stopped. We do the same thing for HA reasons, but just do it a little closer to the DB itself (so you get the extra tracker capacity).
-Dormando
