Service-oriented architecture (SOA) is a highly promising technology for software development. It can greatly reduce the development time and cost and achieve system agility. Web service is the current practice for realizing SOA. Due to the nature of widely distributed web service providers, the service performance could be impacted when the network traffic is congested. There are methods, such as communication cost based service selection strategies, developed to cope with the problem. In this paper, we propose a complementing technology, migratable web services, to overcome the performance problems for real-time applications. A service can be implemented by multiple light weighted web services and some of them can be migrated to strategic sites to minimize the communication delays so the real-time requirements can be met. We have designed an infrastructure to support web service migration and execution. Also, appropriate decision mechanism, which decides whether a web service should be migrated and where to send it to, is the key for major performance gains. We use genetic algorithm to efficiently make the best migration decisions. Experimental studies show that the approach can achieve significant performance improvement.