MC-EZBC is one kind of subband wavelet coders exploiting temporal correlation and fully embedded in quality/bit-rate, spatial, and frame rate. By combining MCTF and EZBC and other non-trivial techniques as well, the latest MC-EZBC achieves the best rate-distortion performance in literature. However, the way to find the /spl lambda//sub mv.opt/, used for optimally pruning the motion vector trees still allows room for improvement, since it is not adaptive to the motion vector trees and it costs high off-line computation as well. We propose a novel method to find an adaptive /spl lambda//sub mv.opt/ for each motion vector tree. Simulation results show that the new way is fast and efficient, and that it brings a gain about 0.05-0.2 dB for widely used testing video clips.