Web data extraction is concerned, among other things, with routine data accessing and downloading from continuously-updated dynamic Web pages. There is a relevant trade-off between the rate at which the external Web sites are accessed and the computational burden on the accessing client. We address the problem by proposing a predictive model, typical of the Operating Systems literature, of the rate-of-update of each Web source. The presented model has been implemented into a new version of the Dynamo project: a middleware that assists in generating informative RSS feeds out of traditional HTML Web sites. To be effective, i.e., make RSS feeds be timely and informative and to be scalable, Dynamo needs a careful tuning and customization of its polling policies, which are described in detail.