ga and multi-start technique

Hi, My optimization problem is nonlinear, nonconvex, mixed integer, and with a wide search space ... so I am stuck with ga ... my nvars = 126 ... I've been told that the PopulationSize should be at least twice the nvars ... if I went with PopulationSize = 250~300 it would be very slow also it might fell in a local minimum as it has one start in such wide search space ... I am thinking to apply the multi-start technique (I know that it is not available for ga and I don't know why) I am thinking to run like 100 parallel runs, each with PopulationSize = 50 for example ... then I will collect the few fittest individual from each of the 100 runs and combine them in a strong/fit InitialPopulation for one last run ... do you have any better suggestion ? how to separate/store say the fittest three individuals at the end of each 100 runs ? thanks in advance,