It reveals that it is possible to successfully set value for this future, but success callback will not be executed and failure callback instead will be called. As in previous case it is the case of concurrently setting and cancelling future under race.
Good catch! I've revised our SettableListenableFuture implementation once more, now centralizing all state into the inner FutureTask subclass (through overriding FutureTask.set and FutureTask.setException and bypassing FutureTask.run completely).
I'll ping you once this is available in a new 4.3.7 snapshot.