Don't confuse the issue by pretending it's all about collaboration and economics of software. It doesn't make sense to try to shoehorn my software idea into an existing framework exclusively due to price and availability. Just because there's a square peg available for free doesn't mean that it'll fit a round problem, even if a square solution may take longer.
I predominately work in computational analysis and have spent a significant portion of my career trying to figure out physical problems (first in video games and now in engineering analysis), particularly in the finite element/CFD domain. That makes OpenFOAM is a classic example for me -- it's the benchmark for open source CFD analysis. But I'm still employed at an engineering firm developing our own numerical analysis tools.
OpenFOAM is quite good at a very small subset of what it claims to do, but it doesn't do *everything* well. Unfortunately, the framework is sufficiently mature at this point that trying to fork it and address those flaws would be a colossal undertaking. This means that for many toolsets, starting from the ground up is simply a more attractive alternative. Could we reuse a few elements deep in the integrators? Maybe, but those would come with their own baggage.