I'm a business analyst by trade and I say increase your sample space.

That is, measure iterations over a longer period of time and either track CPU usage along with that time (allows you to pick a sample space with the minimum of CPU usage spikes) or do your best to ensure the thread is not interrupted excessively by the OS. (set the thread to real-time priority if you must)

Then, count the loops performed in say, 30 seconds then finally divide it down so you know the average processing time in MS for a single pass.

That average will become increasingly more accurate the more time you sample, as long as you've taken care to avoid excessive thread interruption, you'll find that figure to be suitable for your purposes.