Deactivate LTM Pool Members for generic TCP services based on performance hit?

With HTTP services there are plenty of ways to take pool members out of service, or flavor load balancing decisions using sophisticated load balancing algorithms on the LTM, but for a service like Telnet or SSH, we find that in a multi-server pool, a server can become overburdened (high CPU or low memory) that causes the application to be sluggish or non-responsive, yet the LTM has no way of knowing it because a simple check that the IP stack is answering on port TCP/22 isn’t failing. If this were an HTTP service, a PREDICTICE or FASTEST load balancing method could have the intelligence to detect the server is sick, but for a Telnet or SSH service, what can be done to detect application not responsive, yet IP stack is? We use TCP-Half-Open monitors so that the LTM’s monitors don’t leave tons’ of open connections on the server, otherwise the monitors themselves will cause our applications problems. Any suggestions short of re-writing our applications?