Faster in Python 2 than Python 3 - why?

def dump(a):
    # print("Array: ",",".join((str(i) for i in a)));
    pass

swaps=0
def swap(a,i,j):
    global swaps
    tmp=a[i]; a[i]=a[j]; a[j]=tmp;
    # print("Swap:  "+"   "*i+"^^ "+"   "*(j-i-1)+"^^\n");
    swaps+=1;

import random
import time
sz=1000000; top=sz*10-1
start=time.time();
# a=[random.randint(0,top) for i in range(sz)]
a=[int(random.random()*top) for i in range(sz)]
# a=[6,5,3,1,8,7,2,4]
print("Build array: ",time.time()-start)

dump(a);
start=time.time();
# Build the heap
for i in range(len(a)):
    # dump(a);
    # print("Ins:   "+"   "*i+"^^")
    x=i;
    j=(i-1)//2;
    while (j>0):
        # print("Cmp:   "+"   "*j+"^^\n");
        if (a[x]>a[j]): swap(a,j,x); x=j
        j=(j-1)//2
    if (a[x]>a[0]): swap(a,0,x);
print("--- Heap built (%d swaps) ---"%swaps);
dump(a);

top=len(a);
while (top>1):
    # print("Remove %d from heap"%a[0]);
    top-=1
    swap(a,0,top);
    par=0;
    child=par*2+1;
    while (child<top):
        swapme=par;
        # print("Cmp:   "+"   "*par+"^^ "+"   "*(child-par-1)+"^^ ^^");
        if (a[child]>a[par]): swapme=child;
        if (child<top-1 and a[child+1]>a[swapme]): swapme=child+1;
        if (swapme==par): break; # Parent already greater than both children
        swap(a,par,swapme);
        # dump(a);
        par=swapme;
        child=par*2+1;
print("--- Array sorted in %d swaps total ---"%swaps);
print("Total time: ",time.time()-start);
dump(a);