There are two flaws in this.
a) Performance of people need not follow 'normal distribution'. Normal distribution occurs when the variation is due to randomness (or chance). Consider two scenarios - one person taking the same test (rather similar test with same pattern) multiple times, and different people takes the same test. One person takes same test multiple times follows normal distribution, as the variation is due to randomness. However different people taking same test may not follow normal distribution. When different people takes same test, the variations are due to randomness as well as difference in individual's skill. Variation in skill need not follow normal distribution.
b) According to statistics, when we take samples out of a population (eg: depts of 100 people in an organization of 10000 people), the mean (average performance) of the samples (individual depts) should follow normal distribution. However current practice is to enforce individual depts normally distributed with same means. This is mainly because no dept head will be ready to admit that his dept performance mean is less than company mean. That simply means the performce ratings are force fit rather than actual performance.