Invalid block count in EvictedTupleAccessException

[apavlo]

We are getting random a EvictedTupleAccessException where the block count is negative:

java.lang.AssertionError: Unexpected non-negative block count '-30138'
        at org.voltdb.exceptions.EvictedTupleAccessException.<init>(EvictedTupleAccessException.java:35)
        at org.voltdb.exceptions.SerializableException$SerializableExceptions$5.deserializeException(SerializableException.java:79)
        at org.voltdb.exceptions.SerializableException.deserializeFromBuffer(SerializableException.java:220)
        at org.voltdb.jni.ExecutionEngineJNI.throwExceptionForError(ExecutionEngineJNI.java:169)
        at org.voltdb.jni.ExecutionEngine.checkErrorCode(ExecutionEngine.java:101)
        at org.voltdb.jni.ExecutionEngineJNI.executeQueryPlanFragmentsAndGetDependencySet(ExecutionEngineJNI.java:390)
        at edu.brown.hstore.PartitionExecutor.executePlanFragments(PartitionExecutor.java:3231)
        at edu.brown.hstore.PartitionExecutor.executeLocalPlan(PartitionExecutor.java:3044)
        at edu.brown.hstore.PartitionExecutor.executeSQLStmtBatch(PartitionExecutor.java:3487)
        at org.voltdb.VoltProcedure.executeQueriesInABatch(VoltProcedure.java:1161)
        at org.voltdb.VoltProcedure.voltExecuteSQL(VoltProcedure.java:1087)
        at org.voltdb.VoltProcedure.voltExecuteSQL(VoltProcedure.java:1058)
        at org.voltdb.benchmark.tpcc.procedures.slev.run(slev.java:96)
        at sun.reflect.GeneratedMethodAccessor7.invoke(Unknown Source)
        at sun.reflect.DelegatingMethodAccessorImpl.invoke(DelegatingMethodAccessorImpl.java:43)
        at java.lang.reflect.Method.invoke(Method.java:606)
        at org.voltdb.VoltProcedure.call(VoltProcedure.java:600)
        at edu.brown.hstore.PartitionExecutor.executeTransaction(PartitionExecutor.java:2492)
        at edu.brown.hstore.PartitionExecutor.processInternalTxnMessage(PartitionExecutor.java:1415)
        at edu.brown.hstore.PartitionExecutor.run(PartitionExecutor.java:1170)

I thought it was because the EE was overwriting the result buffer in the AntiCacheManager thread, but it appears that I was wrong.