It seems to work. After the system file is processed, there is hardly any overhead for producing small programs which use the system.
So, I am now building a linker and getting ready for a full compile.
Sometimes modularity really works. I only need one run of the compiler to dump object files from the AST; after that, it's just writing and debugging the linker which produces the C code. So, the three hours to produce the object files are now mostly irrelevant.