This study (which HughRistik originally pointed to here) suggests that IQ distribution might be better modeled as two overlapping normal distributions, one for people who are not suffering from any conditions disrupting normal intelligence development (such as disease, nutritional problems, maternal drug or alcohol use during pregnancy, etc.) and the other for those who suffered developmental impairment. If this model has some validity the Flynn effect could perhaps be explained as a reduction in the number of people falling into the 'impaired' distribution due to improved health and nutrition in the population. This would seem to explain an increase in the average score without a corresponding increase in the number of 'geniuses'.