In balanced load allocation problem in complex systems like large-scale data center and P2P overlay network, the various load allocation methods is proposed. In order to achieve better balanced load index, many methods, however, are in pursuit of more and more complicated algorithms, which makes the time and algorithm complexity hard to control. Based on the study of the original balls-into-bins model, Azar balls-into-bins model and balls into non-uniform bins model, the paper brings forward an efficient and balanced non-uniform balls-into-bins load allocation model, which is provided with heterogeneous balls, heterogeneous bins and different bin selection probabilities. The model can achieve rational largest load with high probabilities, at the cost of little time and algorithm complexity. The model is extensible and can be applied in many domains.