He doesn't give a lot of details, but somewhere in the first 20 minutes of this talk (probably 15 minutes in), Cliff Click says that Hotspot had to create fair locks in user-space because kernel provided locks weren't adequate.
Its funny that he throws out the "painful" learning of "don't trust the OS." I don't disagree, but it seems that the main learning should be "know your assumptions."
Consider, this just moved up the stack. A good mantra for developers on the JVM is "don't trust the VM."
Obviously, at some level you are trusting them, but you should also have measurements in place that would let you know which key assumption is failing. Yes, this is easier said than done.
https://youtu.be/Hqw57GJSrac?t=341