Vega-Lite and Falcon author here. Many people ask for scalable visualizations and I think declarative visualization approaches are really promising since systems can transparently optimize data movement and rendering. I have done some more experiments to scale Vega such as https://github.com/vega/scalable-vega but I think what we need is a system that automatically offloads heavy computations to a backend such as https://github.com/omnisci/jupyterlab-omnisci.