In general, 40 samples would certainly not be enough for the experiment result to have statistical significance but perhaps in this case when the code being tes

In general, 40 samples would certainly not be enough for the experiment result to have statistical significance but perhaps in this case when the code being tested is very tight, e.g. just several tens of instructions, it may be enough. Code in this example is testing the CPU compute capabilities and the only source of variance is the CPU internals itself.