It absolutely would depend on architecture, but if the metric is reduced to the mean number of cycles per instruction, any architectural benefits will be implicit in that number. Then all one needs is the clock speed and one can calculate MIPS.

But Maarten raises a good point, which emphasises your point about architecture. If one processor can execute in a single instruction, by virtue perhaps of specialized hardware, a task that requires three instructions on another processor, it doesn't matter if the second processor is performing twice the MIPS. It still takes the second processor longer to execute the same task than the "slower" processor.

DRAT!!! Erik was right again! Don't you hate it when that happems. I guess it really does depend on which processor is most suited to a particular task.

(P.S. Erik, The misspelling is for you. ;-))