The JavaScript language was originally designed to help writing small scripts adding dynamism in web pages. It is now widely used, both on the server and client sides, also for programs requiring intensive computations. Some examples are video game engines and image processing applications. This work focuses on improving performance for this kind of programs. Because JavaScript is a dynamic language, a JavaScript program cannot be compiled efficiently to native code. For achieving good performance on such dynamic programs, the common implementation strategy is to have several layers handling the JavaScript code, starting from interpretation , up to aggressive just-in-time compilation. Nevertheless , all existing implementations execute JavaScript functions using a single thread. In this work we propose to use the polyhedral model in the just-in-time compilation layer to parallelize compute-intensive programs that include loop nests. We highlight what are the scientific challenges, resulting from the dynamism of the language, for integrating automatic polyhedral optimization. We then show how to solve these challenges in the JavaScriptCore implementation of Apple.