The dynamic binary translation(DBT) is used to support legacy binary code, provides program instrumentation, improves code performance, and so on. The dynamic nature of the translation usually incurs extra substantial execution overhead in DBT systems. As a result, DBT may be impractically slow if it do not or can not employ optimizations to offset the overhead. This paper investigates a few optimizations to alleviate the overhead in DBT. We evaluate these optimizations in CrossBit, which is a resourceable and retargetable dynamic binary translator, including block linking, condition codes optimization, register mapping optimization, static-integrated optimization, multithreaded optimization. The experimental results show that these optimizations can improve DBT performance significantly.