if a test fails and didn't clean up resources properly (in this case destroy the cache managers) then next test running on the same thread will see the already existing cluster, and blockUntil.. methods will fail.
Make the fwk report that a cluster is not properly shutdown by a test, when another tries to create a cluster.