Task migration allows a running program to continue its execution in a different destination environment. Increasingly, execution environments are defined by combinations of cultural and technological constraints, affecting the choice of host language, libraries and tools. A compiler supporting multiple target environments and task migration must be able to serialize continuations and then deserialize and continue their execution, ideally, even if the language of the destination environment is different. In this paper, we propose a compilation approach based on a virtual machine that strikes a balance between implementation portability and efficiency. We explain its implementation within a Scheme compiler targetting JavaScript, PHP, Python, Ruby and Java - some of the most popular host languages for web applications. As our experiments show, this approach compares well with other Scheme compilers targetting high-level languages in terms of execution speed, being sometimes up to 3 orders of magnitude faster.