The best part about MSB math, to me, is that at least you get a close approximation to the answer quickly. It's like successive approximations, the errors from making a mistake go down as you move along if you don't need a precise result.

For instance, what's pi * e? Good luck calculating that from the LSB... ;-)

But with MSB, you know it's about 9 because pi is on the low end of 3-4 and e is on the high end of 2-3.

So you can do

3.14 * 2.72 33 + .143 - .28*3 - something small (idea here is that the cross term is small, three sig figs) 9 + .4ish - .9ish 8.5ish

Actual result is 8.5397.

So you can get a few sig figs by saying "eh, these digits are too small to matter" and ignoring cross multiplication of the LSBs for problems that aren't for the purpose of arithmetic.

I use this often in engineering estimation on the fly, because so many constants are only known to 1 sig-fig anyway that it's really hardly worth worrying about more accuracy. Particularly when you're building in a safety factor anyway.