While various optimization techniques have been used in existing thin client systems to minimize network traffic, the screen updates triggered by many user operations will still result in long interactive latencies in many contemporary network environments. Long interactive latencies are unfavorable to users' perception of graphical interfaces and visual contents. Long latencies arise when data spikes need to be transferred over a network while the available bandwidth is limited. These data spikes are composed of a large amount of screen update data produced in a very short time. In this paper, we propose a hybrid cache-compression scheme, called DSRS, to reduce data spikes. DSRS caches the screen updates in data spikes on both server and client sides, and uses the cached data as history to better compress the recurrent screen updates in possible data spikes. Our experimental results show that, with a cache of 512K bytes, this cache scheme can reduce 4.7% - 26.7% data spikes, 5.9% - 15.8% network traffic, and up to 28.1% total latency for profiled Microsoft Terminal Service screen updates. Most of the latency reduction is achieved by reducing relatively long latencies.