Pretty accurate actually, It's just basic statistics. You disregard outliers as anomalies and look at trends. Over the course of a few weeks you get a good picture of how customers see your employees, if you have a 3* average across your staff you cut back the hours on the staff that average below your base average, and watch them more closely. Workload may be too much, they may not be a good fit, they may need more supervision etc. etc. Sure it messes with peoples lives, by making them more directly accountable for their performance.
A poor experience drives customers away, no customers means the restaurant goes under, restaurant going under means entire staff is unemployed. So yeah, get rid of crappy servers.