They should not really, it is possible you are talkign about a Vector that only grabs some functions, the rest it may pass to the orriginal BIOS. This would add a few extra instructions when decoding the function.
As many people has said, it is a lot of effort to remove a few cycles in some code.
However I agree that some functions can benifit from ASM code, but these are normally just screen drawing, etc.
The effort is better put towards a hardware solution. Over the years many of the TOS and (X)BIOS functions have been hacked and changed about to give better performance, there is little more that can be done.
There are just limits to how much code can be optimised. Remember the OS does not contain the code for the APPs, if your APP is 1Mb then there is 1Mb of code that may or may not be well writen.