One way to look at is that local search does exploitation, and the genetic algorithm does exploration. To increase exploration, you could make the search operator (e.g. crossover) more random. In fact random restart of the local optimization technique should be a baseline to measure the effectiveness of any memetic approach. To reduce exploration, you might want to minimize the disruption of the schema targeted by the local optimizer. Here's a paper where I discuss what I mean by that:

https://www.researchgate.net/publication/2274675_Putting_the_Genetics_back_into_Genetic_Algorithms_(Reconsidering_the_Role_of_Crossover_in_Hybrid_Operators)?ev=prf_pub

Alternatively, you might consider reducing the exploitiveness of local search...by not finishing the procedure. Here are some interesting ideas along those lines:

http://ieeexplore.ieee.org/xpl/articleDetails.jsp?tp=&arnumber=6256454

Chapter Putting the "Genetics" back into Genetic Algorithms (Reconsi...

Please take a look at the following paper, it should provide you with some good ideas on how to balance the genetic and memetic search, via adaptation and other statistical means:

X. S. Chen, Y. S. Ong, M. H. Lim and K. C. Tan, "A Multi-Facet Survey on Memetic Computation", IEEE Transactions on Evolutionary Computation, Vol. 15, No. 5, pp. 591-607, 2011.

A very interesting topic. Some of the intuition on the role of local vs global has been given in the first answer, in terms of exploitation vs exploration. The features of the search landscape are crucial to determine the best balance, which will be problem-dependent. For a good analysis of the topic, I can recommend the following two references:

Dirk Sudholt:

Parametrization and Balancing Local and Global Search. Handbook of Memetic Algorithms 2012: 55-72 - https://www.researchgate.net/publication/51940972_Memetic_Algorithms_Parametrization_and_Balancing_Local_and_GlobalSearch

Dirk Sudholt:

The impact of parametrization in memetic evolutionary algorithms. Theor. Comput. Sci. 410(26): 2511-2528 (2009) - http://dx.doi.org/10.1016/j.tcs.2009.03.003

Article Memetic Algorithms: Parametrization and Balancing Local and ...

Thank you all 

now I wonder if we can come up with a method that can learn (or even predict) the search landscape!!

Ahmad, there are many methods for fitness landscape analysis. I suggest you take a look at:

Peter Merz:

Memetic Algorithms and Fitness Landscapes in Combinatorial Optimization. Handbook of Memetic Algorithms 2012: 95-119

https://www.researchgate.net/publication/278698396_Memetic_Algorithms_and_Fitness_Landscapes_in_Combinatorial_Optimization

for a specific treatment of the topic regarding memetic algorithms, and follow the references therein for related work.

Chapter Memetic Algorithms and Fitness Landscapes in Combinatorial O...

Thanks Carlos