If I understood correctly then MIPS is measured based on the capacity of mainframe machine to process no of instructions per second and it is then fixed for all the tasks performed on that machine
You misunderstand. IBM used
to give out MIPS ratings on their machines when they were announced to be available, but stopped doing so because MIPS was a misleading number -- with instruction caches, multiple processors per machine, and so forth a MIPS number could be measured many different ways, giving many different values, and all could be considered "correct" for some value of "correct". They now announce MSU (million service units) for each machine. This value is not calculated nor measured -- it is provided by the vendor as a way to measure relative throughput of two machines.
If you are measuring a number, it is not MIPS as that term is generally used in the industry.