You may also could play with the aspect ratio. Instead of 1:1.35 you could use 1:78 or 1:2. This way you can have resolutions of ~340x600 (close to the monitor standards like HD) or 320x640 (same relation as L5 screen) with same performance as right now.
However optimum would be 360x720@30fps (a quarter of L5 physical screen resolution, identical with default scaling), but this needs 25% more power, which is 7% above your pipeline. 360p@30fps could be possible, which is nice to upload videos to video platforms.
Are you speaking about 2x res like 390x520 to 780x1040 or like 200k px to 400k px? If the first is the case, it’s 4x and 16x res what you described (which makes more sense to me when looking at FPS numbers).
I usually use “resolution” as synonym to total pixel count, because it makes more sense to me when thinking about computation power (even if 4 times pixel count does not mean 4 times computation time).
I did not speak about scaling the image in that way, but about ignoring more pixels from two borders. The camera takes 13MP, the pipeline cuts off let’s randomly say 3MP (maybe driver has to do this? the aspect ratio has to be like the final image) and the rest of the raw image will be combined to the let’s say 360p video-image.