Resolution:
With LiveCluster 3.1.1, there is a known issue when a Service is cancelled. Any Engine currently working on that Service will not receive the notification to restart as it should. Rather, it will continue working on that task until it is completed.

Service Pack 1 for LiveCluster 3.1.1 corrects this issue. To obtain the Service Pack, go to http://customer.datasynapse.com, go to Releases, then 3.1.1.0. The Service Pack is a JAR file, and the readme.txt file explains the installation

For an immediate workaround, kill the Engines that appear to be available, but are not taking tasks.

In LiveCluster 3.1.1 (even with Service Pack 1,) it can also appear that cancelled Tasks don't cancel. This happens when Tasks take less then 30 seconds. Engines don't get messages from the server except when they report in, which is on each heartbeat or when they're done with a task. Heartbeats happen about every 30 secs. So with short Tasks, Engines do not discover the cancellation until Tasks are finished. Using LiveCluster 3.1.1, you should not expect immediate cancellation of your Tasks - it will take at least 30 seconds.

In GridServer 3.2, Engines receive cancellation notifications immediately through their fileserver, fixing this issue.

After cancelling Services, Engines fail to restart (LiveCluster 3.1)