What is the best way to solve A\b using distributed computing and sparse matrix?

I´m trying to solve a large linear system [A]{x}={B} using the traditional backslash operator and distributed computing. However results are exactly on the other hand of what I expected. An increment in the number of workers is making solution slower. Matrix A is sparse with 1e4 order. What am I doing wrong?

n=1e4
A=sprand(n,n,5e-2)
B=rand(n,1);
[l,m,v]=find(A);
v=distributed(v);
spmd
   C = sparse(l,m,v);    % distributed sparse array
end

tic
sol_01=A\B;              % solution using local array
toc

tic
spmd
  sol_02=C\B;            % solution using distributed array
end
toc

sol_01 is many times faster than sol_02. Some examples:

- elapsed time for sol_01 = 21.38s
- elapsed time for sol_02 using 4 local workers = 40.66s
- elapsed time for sol_02 using 4 local + 4 remote workers = 43.38s

The current setup employs a 1 gigabit router (Asus N66U) and has been tested to speeds up to 100 megabyte/s. Windows task manager shows activity below 80% on the network. Both machines share the same specs: i7 4790 - 4 cores, gigabit adapters.

[Paulo Ribeiro]

I wonder if this is a result of using 2 quad core nodes as a total of 8 workers. For these intensive linear algebra computations would it be better 2 workers only, with 4 cores each?