The evolution of multimedia document production and diffusion technologies have lead to a significant spread of knowledge in form of pictures and recordings. However, scholarly reading tasks are still principally performed on textual contents. We argue that this is due to a lack of semantic and structured tools: (1) to handle the wide spectrum of interpretive operations involved by the polymorphous intensive reading process; (2) to perform these operations on a heterogeneous multimedia corpus. This firstly calls for identifying fundamental document requirements for such reading practices. Then, we present a flexible model and a software environment which enable the reader to structure, annotate, link, fragment, compare, freely organize and spatially lay out documents, and to prepare the writing of their critical comment. We eventually discuss experiments with humanities scholars, and explore new academic reading practices which take advantage of document engineering principles such as multimedia document structuration, publication or sharing.