I switched from old sysbench to modern sysbench (1.0) last year. I use sysbench a lot and have never set autocommit=0 in my.cnf. I also have never had this problem. I also know that correlation != causation. I have had problems in the past while modifying the insert bench when autocommit was set wrong. 

I think you mentioned something about running sysbench on s390. I would love to read more about that.