The default number of non-blocking threads is available processors * 2, which is too large considering that the tasks running on non-blocking threads are supposed to be CPU-bound. We could use available processors + 1 to avoid special cases when running on 1 vCPU.

The test suite already uses only 6 blocking and 6 non-blocking threads (TestCacheManagerFactory.NAMED_EXECUTORS_THREADS_WITH_QUEUE), but it doesn't set the number of available processors. Code that spawns both blocking and non-blocking tasks based on the number of available processors may fill both thread pools, making BlockingRejectedExecutionHandler throw a CacheBackpressureFullException. We should set -Dinfinispan.activeprocessorcount=4 in pom.xml in order to match the thread pool sizes.