Hi,
The failover time is mostly determined by how quickly the application you're clustering can stop and start, not really by the cluster. You would have seen that there are multiple resources in the dependency tree, each ones takes a certain amount of time to online/offline. With resources like IP its usually quite quick, with mounts it might take a bit longer and would depend on the applications locks being cleared from the mount, for the application it purely depends on the applications so we can't do much about it.
In the recent versions of VCS/Infoscale their is a feature called Intelligent Monitoring Framework or Asynchronous monitoring Framework that detects state changes by talking to the kernel directly. In failures and online/offline operations this could give you a few seconds improvement here in there as you don't need to wait for polling to occur (legacy method of checking resource states).
HTH
