In this paper, we study the problem of allocating concurrent jobs composed of situated tasks, underlying the distributed deployment of the MapReduce design pattern on a cluster. In order to implement our multi-agent strategy which aims at minimising the mean flowtime of jobs, we propose a modular agent architecture that allows the concurrency of negotiation and consumption. Our experiments show that our reallocation strategy, when executed continuously during the consumption process: (1) improves the flowtime; (2) does not penalise the consumption; (3) is robust against execution hazards.