Hi all!
Here a like-OS UEFI application is used.
This is an extreme method to remove several noises in benchmark.

This is yours opcodes and CPU, nothing else.
Any variance is related to variations in cpu frequency or in to acces to clock (I think,

).
Implementation of Paoloni method under Windows show unreliable result, but consistents (see Vars of Vars

):
Spurious = 2 [-] Number of Spurious min. Values
MaxMaxDiff = 87028 [Tick] Max. Spread of the Values
MeanVar = 1.924E+0005 [Tick^2] Mean Variance
VarOfVars = 5.433E+0010 [Tick^4] Variance of Variances
MinsMSE = 8.793E+0001 [Tick^2] Linear Regression MSE
R2 = 9.995E-0001 [-] Determination Coeficient
Overhead = 87.3244 [Tick] Measurement Overhead
Measure = 15.0080 [Tick] Code Execution Time
(I changed this Jan 19, 2023 because results were for 50000 repetions instead of 10000, were better result but a lot more variance)
But true values for same machine are others (and are very close to what you obtain when running UEFI BootServices):