How to improve the speed of this parfor loop?

I currently have the following code as part of a function in a parallel for loop:

parfor k = 1:K(v)
  [D3_temp(k,:), phy1_temp(k,:), phy2_temp(k,:), phy3_temp(k,:), phy4_temp(k,:)] = ...
  recursions(R_test, R, Nv, K, m, v, k, D);
end

where m, v are outer loop indices. Nv, K are relatively small arrays. D, R_test, R are huge arrays and accessed in a complicated manner. recursions is a local function I had to write to allow the code to run with parfor. The outputs of recursions are not initialized outside the parfor loop because I found I could improve the speed by not initializing them. The inputs of recursions are only read from, never written to within the function.

I had to write the local function recursions because parfor would not allow me to access D, R_test, R in the way I wanted. I tried using a nested function, but MATLAB said it would not allow nested functions within parfor. The typical value for K(v) is 100.

Currently the code takes a lot less time to execute when I replace parfor with 'for', but it only uses 1 core for this. My computer has a 3 GHz octa-core processor with 32 GB RAM. I have set my parallel profile to use 7 workers.

Can anyone help me improve the speed? Is parfor the best solution or is it better to use something else from the parallel computing toolbox?