simplifying ^ fails, on exact numerical constants in Mma 5.0



Consider
f[x_, y_, n_] :=  (x^n)^(y/n)

For x, y, positive rational numbers and n an integer
this should compute exactly the same as x^y.

And indeed this seems to be the case

for
f(1/4+1/10^18,  1/2,  n)

when n is 1,2,3 or 4.  But it comes up with a different
answer when n is 5 or more  (using Mma 5.0).

Are the answers different?  N[%-%%,1000] checks numerical
equality, but this gives an meprec error....

Simplify[..] does not simplify to zero.

FullSimplify does better, if you are willing to wait
long enough, (or n is small enough) and returns 0.

  Can a CAS do this right and fast?

RJF