Your question asks "how" the clustering into a normal distribution should be done. A more important questions is: Should clustering into a forced normal distribution be done at all? The first question is based on the assumed "fact" that employee performance will be normally distributed with performances varying from -3 to +3 standard deviations across the spectrum. If it is found that performance does not adhere to the normal distribution curve (NDC), then it should be corrected by applying a forced distribution. In the case of performance ratings, it is also assumed that ratings which are not "normally" distributed are somehow the fault of the managers or supervisors who did the ratings, i.e., "easy/lenient" raters or "tough/harsh" raters (or the effect of "halo" in the ratings done). The fact of the matter is that employee performance need not (and probably is not) normally distrubuted. It is much more likely that the tails of the distribution are uneven with a greater bunching on the left (low end) and tail to the right extending much more than the assumed 3 standard deviations. In summary, it is likely that a small proportion of employees contribute a lot more than the rest. Alternatively, a large proportion don't contribute much at all. It is unlikely, however, that the power distribution approach suggested here would be adopted rather than the normal distribution approach. The latter is likely to be regarded as "fairer" and a lot easier to defend because of the axiomatic status of the NDC in management and behavioural sciences.