CPU + GPU rendering: Using different tile sizes for each? (some new ideas)

I know this topic has been brought up in the past. And as the the devs have previously mentioned, having individual tiles sizes for CPU and GPU will introduce problems, so they don't seem to be working on implementing this feature...

But i really want this option because it will potentially speed up render times A LOT. And i know that small tile sizes are now much faster on the GPU in 2.8, but even with this improvement they still remain slower than using larger tiles. Because of this problem, for some people, using only the GPU instead of both the CPU + GPU can still be faster in some situations (depending on hardware).

So here's an idea to help fix one of the the problems with having different tile sizes:

Based on the render output resolution, and the selected CPU tile size, Blender could work out what GPU tile sizes will actually fit in the tile grid (so that no gaps/holes/odd sizes exist), and then creates a list of available GPU tile sizes that fit correctly. The user could then select one of those options.

I'm just basically brainstorming here.. And hoping the devs will eventually come up with something. Because i'm in an odd situation where my two 1080ti GPUs are faster with larger tiles by enough of a margin that they still render everything faster than using both the CPU + GPU in everything i've tested so far. I'd imagine there's many other people in the same boat as me, and if the devs can ever pull this off somehow it will put Blender/Cycles above the vast majority on alternative renderers in regards to render times.