Scientific visualization techniques allow the rendering of complex data with a large panoply of tools, ranging from multi-pass rendering for advanced illumination, direct interaction widgets for object manipulations, multi levels of resolution, object-space or screen space partitioning for fast and parallel interaction. In reality, making all this work in a high-performance environment requires a careful balance of end-to-end components, from disk I/O, to parallel filtering and rendering, to remote image delivery. We will illustrate several example applications and show concrete limitations and bottlenecks of parallel visualization.