This solution works, but as you point out balancing over your load balancers is a huge hack and relies basically on DNS.
If you replaced wackamole/DNS RR with BGPd on your varnish/pound boxes, you would achieve the same goals but be able to fully direct your traffic flows yourself vs. a random RFC-busting DNS cache somewhere.
The other big downside to this solution is being forced to run a l2 broadcast domain for failover to work. Fine at your scale here, but when you get into dozens or hundreds of switches and larger scale I firmly believe (if at all possible) dropping L3 down to the access port is the way to go. Debugging STP issues on such networks is about the last thing in the world I'd like to be doing on a friday night.
If you replaced wackamole/DNS RR with BGPd on your varnish/pound boxes, you would achieve the same goals but be able to fully direct your traffic flows yourself vs. a random RFC-busting DNS cache somewhere.
The other big downside to this solution is being forced to run a l2 broadcast domain for failover to work. Fine at your scale here, but when you get into dozens or hundreds of switches and larger scale I firmly believe (if at all possible) dropping L3 down to the access port is the way to go. Debugging STP issues on such networks is about the last thing in the world I'd like to be doing on a friday night.