So this example shows that the upper limit of refresh rate is defined with MCU productivity and has nothing with limits which come from physical realization of LED device. It is true at least for most C51s (if only you do not use 100MIPS devices of course :)
I never has a problem, even with 33MHz 12 clockers when doing monocolor displays. What you say above is EXACTLY why I had to go Cygnal when I went color. I got 80% of what was required with some fancy speed code in assembler (which now run on the Cygnal - still needed even at full speed) but the last few hurdles I just could not overcome with my old favorite the 668. So Now it is a Cygnal F12x

Erik