Is there a way to maximize the MSX speed?

depends on what kind of speed you have in mind.
afaik turning off screen gives you faster blitter copies but wont affect cpu time at all.

Thanks for some hints but is it possible to get some code which help maximizing the Z80 performance in the MSX? I just need to perform some math calculation as soon as possible.

Make your program for turbo-R, enable the R800. Or work on your algorithms so they are faster.

There are basically two things you can do:

- Disable Interrupts, if your code runs for long periods, might shave off a few time... But let's be realistic, not much, there 60 (or 50) interruptions on a second, even though not really fast, it will not save a lot of time
- Optimize your math code a lot... z80 ALU is not fast (it seems it is internally a 4 bits ALU), so, the basics, use the lowest count of bytes that you need, avoid multiplication and division like hell, if you can use look up tables or similar tricks to accelerate calculation, do it...
- Optimize your ASM code a lot as well... Sometimes you need to sit down with an opcode table and check if the advantage of using indexation is worth the extra cycles those instructions take... Try to use registers as much as you can... sometimes a JP is faster than a JR if the condition is usually met at the first time (JP always takes 11 cycles, no matter what, JR takes 13 cycles if condition is met and 8 if it is not met), it takes an extra byte of memory space though, but if the condition usually is met in the first check, it is 2 cycles.... Also it is wise to make the code in a path that it doesn't branch a lot in the happy day scenario... Branchings should be the exception, not the rule.

But, again, it is difficult to figure out how much you will be able to speed up your code, it will depend upon how optimized it already is, and how much more optimization can be done, but, one thing needs to be said, z80 is not the fastest 8 bit kid around to do math (probably Motorola 6811 is the fastest one)... If you can lower the precision and use look-up tables / pre calculation, that is the best way to speed up things... I remember trying several ways to do CRC routines and for 1024 bytes it was not exactly acceptable in my book, unless using lookup tables...