dual setup engine is automatically used. the tessmark numbers did not increase because of a separate issue: a corner case based on the specific tesselation shaders used in that demo, which should be fixed with the next catalyst.
the dual-dma engine is not enabled yet in opengl, but is planned. it should be quite automatic: upload will always use one dma, download will use the other. the only care is to avoid using blocking API (so ReadPixels should go to a PBO instead of to memory, and the app should check the status with a sync object)