Different web-based Machine Translation systems have their own design in handling the translation task. In particular, we made a careful study in how to better handle many users’ translation requests at the same time, and the problems of client request's starvation and server's bottleneck. We proposed a Client-Server architecture based on Moses Statistical Machine Translation (SMT) toolkit for multi-language translation environment with a better way to handle users' requests. We first created a bilingual corpus by catching bilingual context from the Internet and we made use of this information to be aligned with each other for training of SMT model. In our design, internal queues are used to store users' requests, and a multi-threading approach is applied in controlling user’s translation request. Different experiments are performed and showed that our proposed design can minimize client request's starvation and server's bottleneck, and maximize system's efficiency.