The best way to radically increase the intelligence of humans would be to use Greg Cochran's idea of replacing rare genetic variations with common ones thereby greatly reducing mutational load.  Because of copying errors, new mutations keep getting introduced into populations, but evolutionary selection keeps working to reduce the spread of harmful mutations.  Consequently, if an embryo has a mutation that few other people have it is far more likely that this mutation is harmful than beneficial.  Replacing all rare genetic variations in an embryo with common variations would likely result in the eventual creation of a person much smarter and healthier than has ever existed.  The primary advantage of Cochran's genetic engineering approach is that we can implement it before we learn the genetic basis of human intelligence.  The main technical problem, from what I understand, from implementing this approach is the inability to edit genes with sufficient accuracy, at sufficiently low cost, and with sufficiently low side effects.