Recent advances in just-in-time compilation for JavaScript have made it possible to deploy large-scale applications using the HTML 5 platform. Unfortunately, web applications are still orders of magnitude slower than native applications. We propose collaborative optimization as a way to transparently improve performance by harnessing the "collective knowledge" about how individual web applications run. In essence, we extend traditional profile-guided optimization to collect profiles over many users, summarize profiles at web-scale, and then distribute optimization hints to any users able to take advantage of such hints.