In WT-3512 the use of timestamps in timestamp_abort changed to use a rwlock so that the oldest/stable timestamps did not get used in an incorrect way. That change also changed a simple increment into an atomic increment.

The atomic operation results in 13% fewer operations being completed during the test. If an increment gets lost because it is not atomic and/or two threads use the same value, that is okay and does not affect test correctness. It is beneficial to have more operations in play for this abort test.