Doesn't your question presuppose that the height distribution tables reset every year? If the height distribution is a fixed scale over time, then the bottom 3% of that distribution could easily account for far more than 3% of the children that are currently in that age range.
Late-ish Edit for Clarity: We don't weigh all the children every year to calculate new tables, we have enough data to figure out how much a 'normal' child should weigh. The percentiles are based on normal age / weight / height correlations and not their relative weight in the world's current sample of children.
Late-ish Edit for Clarity: We don't weigh all the children every year to calculate new tables, we have enough data to figure out how much a 'normal' child should weigh. The percentiles are based on normal age / weight / height correlations and not their relative weight in the world's current sample of children.