Tests demonstrated that the wireless link quality of statically deployed sensor networks frequently varied in a short time span. Traditional link estimation methodology works well in the varying case of long time span, but poor in that of short time span. This paper proposes a multi-link cooperative forwarding protocol on fine-grain gradient strategy (MCFS), which can avoid the influence of quality jitters on parts of forwarding links through single-send-multiple-receive plus strategy on the synchronized random contention of ACK. A new channel model was developed on NS2 platform, which could simulate the link quality changing in a short time based on the model of two states non-homogeneous Markov chain. Through this channel model, the following conclusions would be reached: (1) MCFS protocol can adapt and work well with short time varying of link quality; (2) In reasonable configuration, MCFS was quite effective on performances in the network delivery rate and energy efficiency, compared with mono-link optimized protocol by the criteria of HOP/PRR, disjoint and braid multi-path forwarding protocols; (3) The good features of MCFS were independent of both network scale and nodes density of deployment.