Failover strategy for memcached

Since you would be running 20 instances on the same machine I don't see any benefit unless you have different applications that are using memcache and want to separate out the instances for that. Having a single instance versus 20 would shave off the milli/microseconds taken in the hash table lookup to find the correct instance to connect to.

If you are interested in knowing how memcache works then follow the link, I have answered the question before. Would help you make an informed decision.

Yes you can setup your custom application within a Windows Cluster. However that isn't what your user is asking about. Windows clustering isn't an Active/Active cluster configuration. Windows clustering only supports failover not scale out. For a scale out solution where the service runs on multiple servers at the same time that would require using some sort of load balancer in front of the application which would route requests to one of the servers which is running the software application. The load balancer can be a hardware load balancer such as a Cisco, F5, etc. or a software package like Windows Load Balancer service.

All that said...

If you did install your application within a Windows cluster it would remain running on the same physical machine until the software crashed at which point the Windows cluster would attempt to restart it on the same machine, or it would move it to another machine in the cluster depending on how you have it configured. The Windows Cluster service doesn't monitor what your software is doing and your software could reject sessions all day long and the Windows Cluster would consider this totally normal.