I love the current C implementations on the C51. They are doing a great job in an "hostile environment".

I have done most of my C51 work using PLM51, and C is definitely a couple of steps forward.

But when a super-loop starts to be too complicated, I would go for a different processor since the development time/costs starts to increase when switching to assembler. Yes, the call graphs - even if cooperative multitasking - will be so complex that they can't be analyzed. A call tree is limited to the definition of a tree, i.e. a common root that then branches out. But with cooperative multitasking, it will become a directed graph. That is what I meant with the analogy with function pointers. Every switch point can bee seen as a function pointer - a way where a new call tree can start.

This world has seen a large number of architectures since this world has a need of very different applications.

When developing in C, the cost to start using a new architecture isn't so high, counted in time. It may take a lot of money if expensive compilers must be bought. But most architectures has free or low-cost tools available. Not all projects needs the best compiler and/or debugger the market can supply.

With the low cost of todays processors, it can also be meaningful to look at slave processors. Maybe multiple identical processors, but just as well different architectures from different manufacturers. Nothing wrong with combining an ARM and a C51, interfacing them using SPI.