While this may seem self-evident, I'm not immediately convinced. The variation in the middle of the bell curve is much smaller than at the edges. The entire cohort in the 25th-75th percentile range may still be substantially at the same point in the curriculum and capable of being taught in a single class effectively.
IQ test scores are normed to have standard deviation of 15 or 16. If the deviation scale is comparable to the mental_age/chronological_age scale (which is why a standard deviation of 15 or 16 was chosen), then in a class of 6-year-olds, the +1 SD children have a mental age of 6*1.15 = 6.90 and the -1 SD children have a mental age of 6*0.85 = 5.1. So in 1st grade you have a spread of almost 2 years of mental age between +1 SD and -1 SD children, who are certainly not outliers. The within-school standard deviation of IQ may be less than 15, compressing the range, but OTOH chronological ages vary, causing another source of variation. I don't think dividing children into three groups is overdoing things. Elementary schools commonly have 3 or 4 classes per grade.
