Sorry -- my explanation of how flashing a cluster works was somewhat garbled, but the overall implication is the same -- when the same dataset is being clusterized twice, the hidden volume being used may not correspond to the cluster table, and then weird things happen.