An increasingly popular mechanism to carry out dispatching of HTTP requests inside distributed web servers is based on the analysis of request content. This is typically realized in the user space since an implementation in the kernel space may result in difficulties due to the connection oriented nature of TCP. Nevertheless, some kernel-based techniques have been proposed in recent years, such as TCP-splicing, redirect flows and TCP hand-off. Although the last one is the most efficient technique, currently few real implementations exist. The paper presents the implementation of a variant of TCP hand-off, which enables an efficient content-aware scheduling in the kernel space of Linux O.S., and discusses a hybrid scheduling algorithm based on the prediction of the content of disk caches. The paper compares the results obtained by the new algorithm with those obtained by a pure content-aware one and proposes solutions for improving transparency and scalability of TCP hand-off.