There is an optimal range for average server utilization in queueing theory. What is that range? What is the downside (cost) if average utilization is above that range? What is the downside (cost) if average utilization is below that range?

There is an optimal range for average server utilization in queueing theory. What is that range? What is the downside (cost) if average utilization is above that range? What is the downside (cost) if average utilization is below that range?