Question: Processor usage in massive computation

I have a pair of markov matrices that are 250x250 and 500x500 in size. I am raising these matrices to high powers, like 10^17 by 10^12. (I'm using a do loop.) If I set digits:=15, Maple will use 100% of the processor that is available. If I set digits:=16+ Maple will only use 1 thread of my processor. Why? It is the same matrix same program. The only thing that changes is the digits of accuracy. This seems exactly backwards from what it should be. As a result, my AMD9590 processor will run full blast when I choose 15 digits of accuracy and gets the biggest matrix done in 1-2 hours. If i set the accuracy to 17 digits, it takes half a day or more to do the same thing!!! 2 digits of greater accuracy should not take an additional 20 hours!

Thanks