Note that radcan will simplify the expression to 1 because it
simplifies 4^n to 2^(2*n). Unfortunately, this doesn't help for
5^n/4^n... even though limit does just fine on
log(5^n/4^n),logexpand:all.
Someone has just got to read the code with the buglist in hand and see
the problems. I think there is a lot of room for improvement in
limit, especially since Taylor/tlimit solves some of the nastier
problems -- taylor didn't exist when limit was written.
-s