How does a replacement work in case of a machine failure?

A head can easily be replaced by a new one. After the configuration of the target storage, iCAS can resume work immediately.
A similar principal is used for High Availability. In the case of a clustered NAS, this means that all servers access the same storage. If one server fails, its work load will be re-routed to a different server.
If virtual machines (VM) are used, a failure of a VM results in the machine being copied to a different ESX server via VMotion. For systems with internal disks, a re-mirroring of the backup is necessary. The way this is done depends on your SLA.