Allocation failures are normal. They happen when a memory pool fills up. In your case, ParallelGC, they are what triggers a GC.
You had a GC pause of 66ms in which your heap went from 1'085'489K to 562'034K at a total size of 1'908'224K. You should see the same when you run the stress tests on EAP with Java 1.7 and PrallelGC.