Faster processing with parfor loop

I am running a matlab code using parfor and using 2/4 local workers. I would like to speed up the code . Can you please suggest how to use the code efficiently in cluster. An example code is:

parfor k= 3:(NZ-3)
  for j=1:NY
    for i=1:NX
      value_dvy_dx = (27*vy(i,j,k)-27*vy(i-1,j,k)-vy(i+1,j,k)+vy(i-2,j,k)) /DELTAX/24;
      memory_dvy_dx(i,j,k) = b_x(i) * memory_dvy_dx(i,j,k) + a_x(i) * value_dvy_dx;
    end
  end
end

[Walter Roberson]

If vy is an array instead of a function, as it looks to be, then your

for i=1:NX

would lead to indices as low as -1, which is a problem.