This avoids allocated an AtomicInteger for every PTagsFactory.
The concurrency needs are met through an AtomicFieldUpdater.

In benchmarks, the AtomicFieldUpdater appears the same or better with lower memory overhead.