Conclusion
Hardware parity RAID offered performance advantages for both operating systems. As long as recovery from RAID hardware failure is tolerable, it would seem hardware is difficult to beat.
Software methods are not all the same, and depending on the type of I/O prevalent in an environment, software parity RAID can provide adequate performance with efficient storage resiliency. Choosing the right software parity method is key, and this testing has advanced some options over others in my mind.
I would not be willing to join pundits who posit that software parity RAID should be avoided at all costs, or that hardware RAID should be shunned because it is inconvenient. But to be frank, I avoid parity RAID except when limited write performance is both tolerable and justified by the storage cost savings.
I now have a better perspective on how these methods perform. Some of the limitations of software RAID were predictable. But I also wonder at some of the surprising results, like high Ubuntu CPU utilization and the poor hardware sequential read IOPS for Windows.
One option I did not consider was the Intel Rapid Storage Technology RAID, which is implemented as a firmware/software hybrid. I had disappointing results with RST or its predecessor, Matrix RAID, a decade or more ago, but it may be worth revisiting to sate curiosity. [Update: RST is deprecated, so this is not worth looking into for me.]
I invite comments on this post.