I know that more precise computation leads to other results, but this is weird: With 8Bit integer, 16Bit float and 32Bit float screenshots look nearly the same regarding colors and the used color difference key. There are only slight differences.
But with 16Bit linear color and 32Bit linear color the difference key gives compete other results and more important: The contrast and brightness effects and color balance are complete different. Looks like color balance has no impact as well as brightness.
From left to right: 8Bit, 16Bit float linear, 16Bit float