He is trying to get more input, so that his sample size is not just 1.

Regarding your 15cb point improvement at 4.4, try maybe 4.35. Steve at GN often talks about clock stretching, in that 4.4 might actually be slower than 4.35 due to clock stretching.
Overclocking for performance is trying to find the optimum point of performance with the available TDP and architecture.
That does not always mean its the highest frequency which i think is what you might be experiencing.

I suspect a 4.35 or 4.375 will beat the 4.4 run.