I suspect there is no perfect answer to your question, you are going to have to compromise in some way, as I'm sure you are already doing.
I have a similar problem to yours at work. The applications that I work on are written in Python instead of Borland C++, but like your situation, these apps rely on a rather large Visual Studio compiled DLL for some functions.
My method of debugging these applications involves a combination of two debugging strategies: the use of an interactive debugger and the so called "printf" debugging technique.
What I basically do is pick one of the two areas as my main debug focus, and that determines my debugging approach:
If for a given situation I decide that I need to debug the DLL with greater detail, then I work with the VS debugger. I set the executable to run in the DLL project as my python script and that enables full debugging of the DLL code. If I need debugging support from the Python side, then I add print statements. If I need a breakpoint on the Python side to inspect some values, I just print all those values and immediately after call a C++ function that does nothing, but that has a breakpoint set in VS.
When I need to concentrate more on the Python side more I use a Python interactive debugger, but I have VS with the DLL project loaded on the side so that I can quickly add any necessary printfs on the DLL and recompile, so essentially the reverse of the above.
I know it's not the answer you expect, but it is a decent solution in my opinion.