Instead of just looking at neighboring sites, look at sites a distance 
away (this is fast on a hypercube).
If the label is the same at any point, put a connection between the sites. If
there are connections 
, make a connection 
.
Now changes in the labels can be propagated long distances in one iteration.
Expect labeling in 
iterations, each of which takes 
multigrid steps, each taking time 
on hypercube. So time
complexity of 
.