Hi Wahhajs,

This is likely because of our current load balancer which uses data from previous renders to optimise future renders, but has to make a guess the first time you render how to divide the work between the computers you have.

Also, the current system has some overhead in that the render officially can't finish a frame until all the computers have finished working on it. We use tile splitting at the moment which divides up a single frame into tiles which are rendered on your computers, then, the images for those tiles are sent to the main computer. Once Blender has loaded the results, it can move onto the next frame.

Earlier this year we started work on a prototype that dramatically changes all this. We now use frame splitting for animation rendering and a brand new dispatching algorithm that no longer needs to guess from previous frames how to optimise the rendering.

ETA for this new release is October.