Don't know if this is a bug or a feature but i Managed to make CPU+GPU rendering in 2.79b (kinda) work by testing out a stupid idea of mine

Basically you just need 2 network cards (wireless usb adapter also works) in the same PC connected to the same network, just add the IP address of your second NIC in Crowdrender tab and set the compute device for it to CPU while keeping your main compute device (this machine) to CUDA of OpenCL, and that's it. have to manage the tile size for better speeds though, Tile size of 16-32 seems to work great for my Ryzen 1700 with a manually set 8-15% in load balancing

Also manage to make Cuda play nice with OpenCL for once (GTX 1060 6gb + RX470 4gb) using the same trick, did the same by setting Local Renderer (this machine) to Cuda while the slave node (2nd NIC IP) is set to OpenCL, and it seems to work just fine so far

Also tried adding a 3rd NIC just for the hell of it and try to see if i could add the two GPU i had (GTX1060 and RX470) + the CPU, and set GPU1 (GTX1060 ) as local master (this machine) and GPU2 (RX470) to NIC2 and then CPU to NIC3 but it doesn't seem to work, blender would freeze abit then Crowdrender would have connection failed errors on both nodes

I did test out with 3 machines to see if the output was ok if i had 2 network cards in my machine with CPU+GPU then connected to two slave nodes that use GPU compute only and the results looks ok to me (used Splash_fishy_cat, BMW27 and pabellon_barclona Demo for testing). was expecting it to not work (with the Master PC having two IPs and all) but somehow it did, though it may need some further testing