Learning in Strategic Queuing Systems with Small Buffers