I can certainly understand the desire to use a sandboxed environment that would enable you to optimize the performance of your program without external distractions (for example, background anti-virus checks, OS updates, browser page refreshes etc.). These indeed affect the CPU and memory and therefore your program's executing-time variabiity. As mentioned in the other answers here, there are ways to reduce such distractions to a minimum.
But in addition, note that in real life your program will likely encounter exactly those distractions that you are trying to avoid here. Therefore, optimizing your code for the optimal case (minimal distractions) may possibly make the code sub-optimal for the common case (typical distractions). In most likelihood, you'd rather optimize your code for the typical use-case, not a hypothetical unrealistic optimal use-case.
I'm not saying that sandboxing the application during performance-tuning doesn't have a place, I'm just saying that you should be aware that it has a downside.
One way to work-aroud this limitation is to run the same code using the same inputs multiple times in a loop, and then using the average timing of all these runs. In fact, this is what the builtin timeit funcion does. If you run your code enough times, the effect of spurious distractions from external processes will be minimized, and you'd get a more realistic picture of what your end-user will see. By optimizing the code to make this average run-time faster, you're more likely to get the same effect for end users (some say that it's better to reduce the maximal run-time rather than the average or median, but that's a separate discussion outside of the scope here).
