This makes no sense.
On most single-threaded benchmarks the workstation outperforms the Core i7 in the MacBook Air; this is not surprising given the Xeons are at 2.7 GHz and the i7 at 2 GHz. But on multi-threaded code and NIO, the little laptop whups the 12-core workstation.
The multithreading I think I understand. The benchmarks are doing a pretty good job of saturating all the processors, and this particular test (a generic object pooling strategy based off an idea in OpenHFT) intentionally creates thread contention to see how well or badly the code handles the locking. The workstation is dealing with 6x the contention, and performance is degraded from 20 million to 7 million operations / second.
But the NIO code doing materially worse I do not understand. The text parsing cases are all slower, and these are straightforward Java code; I am testing out Richard Hightower's Boon text processing code. Unless the Solaris JDK is not as well optimized in these routines, I can't see why it would be slower. I will be setting up Linux under VirtualBox later so I could try that for comparison.
On the plus side, I learned something about Solaris power management. The powertop command will show you what % of the time you are spending running at different supported frequency scales on your CPU. You can use poweradm to either entirely turn it off, or fine-tune it. I turned it off, re-ran my benchmarks, and got from 46,500,000 ops/second up to 50,200,000 ops/second.
# observe frequency stepping
# disable power management
sudo poweradm set administative-authority=none
# re-enable power management
sudo poweradm set administrative-authority=platform
See How to Save Power on SPARC T5, SPARC M5, and SPARC M6 Servers on Oracle TechNet.