No idea about AMOS, but a common trick is to simply time how long it takes to perform a loop of NOP instructions. Each CPU has different timings for NOP (and the loop timing itself will differ from CPU to CPU but in a quantifiable way) and use that to estimate the speed. You'll have to study the actual performance timings for each CPU to know exactly how long your loop ought to take and time a sufficiently large number of iterations.
I'm sure there are better ways.