The problem is actually the opposite, at this point; Linux has seen tons of effort and been polished to hell, but all the FOSS apps that would be alternatives to proprietary ones haven’t seen nearly so much contribution.

Linux “as a product”—the kernel, the common libraries, plus any random distro’s bits chosen at random—probably has as many (if not more) cumulative man-hours put into it, per year, as the “products” of Windows or macOS have.

But can you say the same about, say, GIMP vs. Photoshop? About Audacity vs. Reason?

Even Blender, which sees tons of work put into it and is sort of a “reference platform” that graphics orgs use to demonstrate new tech (once they decide it’s not valuable enough any more to keep it proprietary), doesn’t have the same power as the proprietary equivalents.

But in many of these cases, I think the problem isn’t so much “FOSS vs. proprietary” as it is “every other solution, vs. the solution hacked on and extended by industry professionals and academics who are trying to get something novel done.” When there is such a solution, nobody else is going to be able to match its features, because its features are commercializations of R&D that may have not even been published outside of its bounds. (See e.g. Photoshop’s implementation of PatchMatch when it was first introduced; or hundreds of the technniques used in Pixar’s Renderman currently.)

In cases where the FOSS solution is the industry workhorse, the FOSS solution ends up being the thing that gets extended. LLVM is FOSS, and is also where research into optimizing compilers happens. R is FOSS, and is also the usual choice of platform to bind to for projects like WordNet, because they want to use R to do analysis against their project’s specialized data structures.

But that just means that for those industries, Linux is the locked-in platform, since it’s where these tools (or more specifically, their ecosystems of libraries and tooling, which may not always be as multiplatform as the runtime itself) run natively and/or best. But this is just as bad a situation, in many ways, as having the “industry workhorse” be a proprietary solution like Photoshop or Mathematica. Either way, you’ve got one system with all the effort concentrated on it, where sloppy algorithms get a pass because they’re the only implementation of said algorithm, and there’s no real competition to force the ecosystem forward. Sure, you could read the source and implement a competitor yourself—but given industry-developer network effects, few people start out hopeful for their prospects when they do so.