This study evaluates a deep reinforcement learning agent against a state-of-the-art heuristic for resource allocation in dynamic multicore elastic optical networks (dynamic MCF-EON), focusing on various multicore fiber architectures. The distance between cores influences inter-core crosstalk (InC-XT), a key parameter. The simulations considered the Eurocore topology, using three-core triangular fiber configurations and hexagonally arranged seven-core fibers. The results show that DRL agents outperform heuristics by an average of 18 % in blocking probability, particularly under specific inter-core distance conditions. This superiority is attributed to the adaptability of DRL agents learned during training. The study suggests that DRL algorithms show promise in addressing resource allocation challenges in MCF-EON networks, even under strict constraints.