Mechanisms such as Java Web Start enable on-thefly downloading and execution of applications installed on remote servers, without the need for having them installed on the local machine. The rapid diffusion of mobile devices (e.g., Personal Device Assistants - PDAs) connected to the Internet make these applications appealing to mobile users. However, in many cases the available bandwidth is limited, and its excessive usage can even be a cost when wireless connections are paid on a Kbyte transfer basis. This paper proposes an approach based on Genetic Algorithms and an environmentthat, based on previous usage information of the application (i.e. scenarios), re-packages it with the objective of limiting amount of resources transmitted for using a set of application features. The paper reports an empirical study on the application of the proposed approach on three medium-sized Java applications.