The shape of the curve physically cannot remain constant. As you go lower, the effects of mutations increase in magnitude, and the distribution moves more mutations into the beneficial category. This has to happen at the physical and biochemical level.
Diminishing returns epistasis, a demonstrably real empirical phenomenon, means that adding one more gene copy when you have few will matter a lot more than if you already have many. Increasing the catalytic rate by some X amount will matter much more as a proportion of total when the rate is low, than if the rate is high. Etc. etc.
Simultaneously, as you move closer to a fitness peak, you get closer to having exhausted the total pool of beneficial mutations, which lowers their subsequent probability of occurrence, which means more mutations shift from beneficial to deleterious. Conversely as you accumulate deleterious mutations and move further down in absolute fitness, now more mutations shift back into the beneficial category.
The scenario that GE entails is physically impossible. It cannot occur in reality. It is based on imagining a fixed spectrum of mutations under all circumstances. It is an imaginary mathematical abstraction that does not reflect the real physical reality we inhabit.
Even if one grants that the current DFE of mutations for most organisms look like you describe it, that would only tell us that most organisms are highly adapted, it would not follow from this that inevitable and indefinite fitness decline is the fate of any form of life, as the demonstrable reality of the physical phenomena I described would come into play.
Here are some references for the reality of these effects:
https://science.sciencemag.org/content/332/6034/1190.abstract
https://science.sciencemag.org/content/332/6034/1193.abstract