As I understand what you’re describing, you could, after blurring the merged layer and changing the blend mode, go back to the levels layer, duplicate it, and move this duplicated layer to the top. Then each levels layer would be independently adjustable.
If you made the merged layer a smart object before blurring it, then you could use the Gaussian blur and have the blur be continuously adjustable too.
I too am curious about the effect you're trying to achieve.