The 90th percentile is a measure of stastical distribution, not unlike the median. The median is the middle value. The median is the value for which 50% of the values were bigger, and 50% smaller. The 90th percentile tells you the value for which 90% of the data points are smaller and 10% are bigger.
90 percentile is average of best response time that 90% of users can get. Just to eliminate bad spikes spoiling the entire average response time and also most SLA will say ok if 90% of users get good response time. All depends on SLA.
90tile is not realistic unless you are considering standard deviation. Typically analyze your raw data look at average and report on any significant spikes during specific times
90 percentile is not realistic, but it is a real value standing at 90th percent when sorted in ascending. Average value can be misleading, like a big spike at 2-3 intervals can skew the avg resp time, but 90th percent is a actual value of response time which is good measure of resp time.