The mean time to repair is the average time required to fix a particular fault.
HA clustering systems use controlled redundancy to minimize mean time to repair by having redundant components which can be automatically switched in during a failure.
However, the underlying fault still remains, and this loss of redundancy also needs to be repaired in a timely fashion, or the cluster remains vulnerable to further faults of the same type.
This information provided courtesy of the Linux-HA project at http://linux-ha.org/