Best practice would be to render the 3D scenes first then import the image sequences to the VSE. However, this limits the flexibility offered in the approach above.
Here I am suggesting a more interactive use of the 3D view so that you can visualise the effect of motion ‘between’ adjacent scene strips. That is you could zoom into one point of an image, then place the following image in the correct place to zoom out again. If you pre-render the animations you’ll loose that ability.
Remember that rendering in OpenGL mode is very fast, so any mistakes are easy to fix.
