It doesn't appear to use multi-core, and if anything, it's slower than a regular for loop for most cases.
X = 1:1000000;
tic
Y1 = sin(X);
toc %0.0202 s
Y2 = zeros(size(X));
tic
for j = 1:numel(X)
Y2(j) = sin(X(j));
end
toc %0.0372 s
Y3 = zeros(size(X));
tic
parfor j = 1:numel(X) %4 cores
Y3(j) = sin(X(j));
end
toc %0.1092 s
tic
Y3 = arrayfun(@sin, X);
toc %0.611 s
To see if it goes out-of-order processing, which is often a sign of multi-threaded application, try this:
arrayfun(@(x) fprintf('%d\n', x), 1:100000);
you'll find the numbers print in order, even though the doc does say don't assume it will.
"You cannot specify the order in which arrayfun calculates the elements of B or rely on them being done in any particular order."
With that said, I'd avoid arrayfun if you want speed and control over your code. Instead, control the parallel processing by explicitly using parallel functions, like parfor.
