2 parfor loops and gathering/distributing cell array

I have a question about timing in the following minimal code. It appears that between the first and second parfor loops, time is spent gathering components of the cell array from the workers and then redistributing them to the workers. Is there any way to avoid this while still using parfor and not spmd/distributed arrays. Also, I need to do this in two stages with two parfor loops, I cannot do it with one parfor loop.

Kcell = cell(numNodes,1);
parfor n = 1:numNodes
    Kcell{n} = rand(3,3*numNodes);
end 
parfor n = 1:numNodes
    Kcell{n}(:,1) = 1; % some other operation on Kcell{n}
end
K = cell2mat(Kcell);
% do something with K like a linear solve