By the way, could you, share us with some examples where you need to control things yourself and where Keil cannot help you, please? Especially I ask for:
I often optimize code for speed/size based on absolute position of variables/arrays, relative position of variables to each other

I can not answer that for (I)DATA, but when working with C I very often use an assembler module to define all XDATA variables.

Why?
Tho locate most frequently addressed XDATA variables in the "internal external" RAM which is accessible at full speed where the "external external" RAM need a strech cycle to meet the timing requirements.

When the above made me make the locator module, I later got the idea to locate buffers and such in a way that made them start on nice even addresses easy to remember when debugging.

Erik