As you probably already know, the MultiLib compile mode allows you to generate a UserLib that does not need subsystems for Unicode, Threadsafe, etc.
The downside to this is that it creates a UserLib that is four times larger than a non-MultiLib one (although it is probably about the same size as a normal UserLib plus all its subsystems).
The reason there is no size difference:
It would not be too difficult to figure out which functions are different in the different compile modes (I am welcome to being proven wrong); however the difficult part is to insert the correct symbol name if it has only a normal version (there are currently 11 sections in Inc_Tailbite.pb that deal with the decoration, each section probably has multiple lines involving the decoration).
If that isn't clear, imagine this: in each of those 11 sections, TailBite would have to scan a whole linked list (which could be large or small, depending on the UserLib size) to find out whether to decorate the symbol (adding _UNICODE, etc) or not. It's just not worth it really.
If you do want to do this, you will need to figure out a method yourself. For example manually diffing the MultiLib generated asm, and replacing all the occurences of the decorated function you want to remove with the non-decorated, edit the .Desc file, LibNameObjFiles.txt, and run the batch file.
Not to mention the symbols in the data section...
If it sounds complicated, that's becasue it is.
If you feel so inclined, you can add the size reducing functionality to TailBite (the sources are easily available). If you want it to be inlcuded in the main sources, it must be either extremely fast, or be optional (i.e. a command line option to enable it).
Oh, and you need to be able to generate a patch file so we can merge it into the sources (we are more than likely to change something while you code it)