In a multicore processor, arbitrating the shared resources so as to ensure predictable latencies for hard real-time tasks is challenging. In [1], we have introduced a two-level bus arbitration scheme that fits the needs of heterogeneous task sets, when some tasks have a higher demand to memory than others. In this paper, we show how this scheme can be used to optimise the overall utilisation of the cores while enforcing the schedulability of the whole task set. Our approach both configures the bus arbiter and maps the tasks onto the cores. Experimental results show that it reduces the global utilisation of the cores compared to the traditional round-robin scheme.