how share data before gpu process them?

Sorry, but I can't make much sense out of most things you write. What I do understand is the following:

- You want to render the scene from four perspectives. You must set up four different render targets and then render the scene to each render target, using a different camera for each one.

- How do you want to present the four render targets when finished? In the end you will have to transfer the four images to the main rendering buffer in some way.

- Sending the entire scene over a network, letting another computer render a perspective, send the results back and all that in the short time you have for each frame is no good idea. Do you really want to write something that can only be used by people who have a suitable network at home? How do you want to deal with different hardware configurations (and the resulting differences in rendering quality) on the computers in the network?

Try to go the other way. Optimize the rendering code so that the GPU is up to the task of rendering your four perspectives. Identify the parts that cost too much time and find a compromise between quality and speed. Too many polygons in the models? Then reduce them. Textures too big? Then use smaller ones and save the GPU some time interpolating them. The shaders are too demanding? Then use other shaders which yield most of the quality with less effort. And optimize the data structures you use to organize your objects with. The more time the CPU needs, the less time is left for the GPU to render.