I slipped a note about the source code in... But on further reflection, I suspect for the best performance some reverse-engineering will still be needed even if all components are open (does the source code really tell you what assembly is generated? Does the assembly really tell you how the CPU will execute it?)
When was the last time you needed to read assembly code to debug a kernel-level issue in FOSS software? There's a lot of software you can use to profile the entire kernel and userspace[0], or explore the stack trace[1] of your software. If you're not using those and instead debugging your software by counting SIMD calls in the Ghidra editor, you're wasting an unconscionable amount of time.
People read through assembly to fix compiler issues, not to rationalize what the kernel is doing. You're right that this skill is still required for "perfect" optimization, but it's not really relevant when comparing kernels.