A Very skilled ASM Programmer can Always make tighter faster code then a HLL, but not write it as fast.

two places where you REALLY can make an impact
a) where your derivative has 'unique' features that the compiler does not take into account. one example would be a "high address for mov a@Ri SFR".
b) by example: you locate a 200 character buffer accessed by the above method a a 256 byte boundary and just ignore testing for 'low address overflow' which no compiler would ignore.

I find that one of the areas where "Very skilled ASM Programmer" can make a REAL impact is in XDATA to XDATA moves (with processing inbetween).

In many projects Most of the code does not need to be fast or tight. The extra 5% or 10% is out wieghed by the time saved.
I totally agree. The one thing I 'push' re asm/C is "choose judiciously" which routines you do in what.
It is very difficult for an Asm lover (Jan?) to decide to go C for a given module (but there will be faster code) when asm is not needed and equally difficult for a C lover to go asm (I can write it so much faster) when asm is advantageous.

Erik