that the first thread to launch managed to commit each time, while the
other 3 threads got nothing but conflict errors.  Added some randomization
to boost the odds of success, and to prevent systematic favoring of any
thread.  That appears to work.