ifort -ipo flag strange behavior - performance

Question

I have the following code for testing the intel mkl DAXPY routine.
program test
implicit none
integer, parameter :: n = 50000000
integer, parameter :: nloop = 100
real(8), dimension(:), allocatable :: a, b
integer start_t, end_t, rate, i
allocate(a(n))
allocate(b(n))
a = 1.0d0
b = 2.0d0
call system_clock(start_t, rate)
do i = 1, nloop
call sumArray(a, b, a, 3.0d0, n)
end do
call system_clock(end_t)
print *, sum(a)
print *, "sumArray time: ", real(end_t-start_t)/real(rate)
a = 1.0d0
b = 2.0d0
call system_clock(start_t, rate)
do i = 1, nloop
call daxpy(n, 3.0d0, b, 1, a, 1)
end do
call system_clock(end_t)
print *, sum(a)
print *, "daxpy time: ", real(end_t-start_t)/real(rate)
a = 1.0d0
b = 2.0d0
call system_clock(start_t, rate)
do i = 1, nloop
a = a + 3.0d0*b
end do
call system_clock(end_t)
print *, sum(a)
print *, "a + 3*b time: ", real(end_t-start_t)/real(rate)
end program test
subroutine sumArray(x, y, z, alfa, n)
implicit none
integer n, i
real(8) x(n), y(n), z(n), alfa
!$OMP PARALLEL DO
do i = 1, n
z(i) = x(i) + alfa*y(i)
end do
!$OMP END PARALLEL DO
end subroutine sumArray
Here, sumArray is a handwritten subroutine with openmp that does something similar to DAXPY.
When I compile the code with ifort test.f90 -o test -O3 -openmp -mkl the results are (aproximately):
sumArray time: 5.7 sec
daxpy time: 5.7 sec
a + 3*b time: 1.9 sec
However, when I compile it with ifort test.f90 -o test -O3 -openmp -mkl -ipo the results for a + 3*b change dramatically:
sumArray time: 5.7 sec
daxpy time: 5.7 sec
a + 3*b time: 9.3 sec
So firstly, why is the naive array sum better than mkl? And what does -ipo have to do with the slowdown of the naive array sum? Also, what bothers me is that when I eliminate the loops, that is, when I just time each operation once, the times are just like the first case divided by 1000 (around 5.7 ms for sumArray and daxpy, 9.3 ms for a + 3*b) regardless of using -ipo. My guess is that something about the naive sum in a loop allows the compiler to optimize further, but the -ipo flag messes up this optimization. Note: I know that -ipo in this case is useless since it is a single file.