When you shutdown a node that owns cluster resources, you receive:
System Process - Lost Delayed-Write Data.
To insure that the surviving node restarts services and aquires the resources, the heartbeat to the surviving node is stopped. If a service takes too long to stop, its' disk resources may be grafted to the surviving node, before the service finishes writing. This is definitely the case if the cluster log contains:
... \[FM\] FmpCleanupGroups: Timed out on the CleanupThread
To work around this issue, use a batch file to perform the shudown. In the batch file:
1. Net Stop services, specially Exchange.
2. Net Stop clussvc.
3. Run Shutdown.exe, from the resource kit.