At the end, a thing is very clear: save from the serial test, none of the examined JVM gives dramatically different results. This is particularly true for 1.7 versions, and it was somewhat expected: as in late 2006 / first half of 2007 Sun open sourced nearly all Java core code, the performance characteristics of the two software stacks has to be somewhat similar.
Still, the 1.6 OpenJDK implementation is quite slow in the serial benchmark, while the new open 1.7 stack solve this issue. The Oracle JVM remain somewhat faster, but the margin is quite small.
Anyway, remember that these are the results of a single benchmark suite, albeit a complete and industry-standard one. In other word, is entirely possible that different workloads will lead to different performance results.
Have a nice day!