You make a very good point, the modern day app footprint is cluttered with all sorts of "Bloatware", the end result is that hardware performance improvements aren't equally matched with end user software experience. Naturally an inverse relationship would emerge between the need to make writing software easier for as many developers as possible, with the need to have the most performant software. This is the case due to the fact that the lower you go in the software development stack the harder and thus less accessible it becomes for people to write software. So what we ideally want is the benefits of high level language like javascript with the performance optimisation of C. There's a lot of people working on this problem, particularly in the context of machine learning, where theres a strong need to suck as much performance out of the hardware and at same time expose these to vast community of developers in form of easy to use python API's. This has also occurred in the embedded world with the rise of Arduino, any chip maker today knows that if they want their device to succeed in the market today it has to provide Arduino compatibility.

That being said you make a good point that if we were to start from the bottom again, we could definitely optimise a lot of the app we use today. Thats one of the things that still excites about the embedded space, there's a lot room to reimagine products from the ground up. I think the worst impact the advent of the modern API universe, is the decline in first principles thinking.