I've got an implementation with one ACE module and two DNS servers behind it. Balancing is even but the "show serverfarm detail" command shows a very high failures level (about 1/4 of the total connections).
Moreover - both servers become unreachable from outside from time to time. I graph server performance with MRTG and get gaps in my graph during the time when the servers are unreachable, but "show serverfarm detail" shows non-zero current connection counters.
How could this behaviour be explained?