A challenging issue many online social media have to deal with is facing egocentric workloads that are very frequent. Such a situation is generally due to the simultaneous access of several users to a small piece of data owned by a user or a few ones. A key example is the number of comments posted on the Manchester United Facebook page after the manager announced his retirement (more than 1 billion comments on the related subjects). Since egocentric workloads are transient, two dimensions must be taken into account to deal with them: (1) the rapidity to react to the peak load and, (2) the lightness of the solution or its low cost. Therefore, the first goal of this paper is to exploit the underlying social structure of online social media to localize from which the peaks take place and to face them in their early stage. The second goal is to combine an elastic approach with a load balancing process to sustain the overall performances while minimizing the required resources. Our solution is evaluated through simulation with SimJava. The obtained results show the soundness of the approach as well as it feasibility.