Thank you Mihail Malostanidis for leaving such an informative and detailed comment. The idea of the post was to reflect the gain and not to quantify the percentage of the gain. For taking 99 percentile , the sample size has to be big enough as our load tester and one service is running in the same VM competing for same resources. This also affects on how much time we give to the jMeter threads to spawn up and start making request.

In any case, your calculations are correct and if one has to quantify the percentage gain, we should take higher percentiles to match end user performance