I am afraid that such approach does not solve the issue with memory, receiving the input data is just one part of the puzzle and it becomes meaningless if the rest of the tasks (handlebars rendering, chrome rendering, etc) do not work the same way, because eventually some of these tasks will require having everything in memory.
we have some ideas to make handlebars to work this way (for the output part, not the input), but as I said it depends on the rest of the tasks that are going to be done, and it does not work for all cases/recipes.
because of these points, receiving the input data this way does not represent an advantage right now. in an ideal world each task supports getting input and producing output in a streaming way but this is not possible so far.
