The tradeoff between diversity, multiplexing, and delay in multihop MIMO relay networks with ARQ is studied, where the random delay is caused by queueing and ARQ retransmission. This leads to an optimal ARQ allocation problem with a per-hop delay or end-to-end delay constraint. The optimal ARQ allocation has to trade off between the ARQ error that the receiver fails to decode in the allocated maximum ARQ rounds and the packet loss due to queueing delay. These two probability of errors are characterized using the diversity-multiplexing- delay tradeoff (DMDT) (without queueing) and the tail probability of random delay derived using large deviation techniques, respectively. Then the optimal ARQ allocation problem can be formulated as a convex optimization problem. We show that the optimal ARQ allocation should balance each link performance as well as avoid significant queue delay, which is also demonstrated by numerical examples.