Just for fun, if your memory can handle it:
A = mean(permute(cell2mat(reshape(M_one,1,1,3,[])),[1,2,4,3]),4)
For matrices of that size, a loop will be most efficient, e.g.:
M_one = arrayfun(@(n)randi(17,5000),1:15,'uni',0); % fake data
C = reshape(M_one,3,[]);
N = size(C,2);
D = cell(1,N); % preallocate
for k = 1:N
D{k} = mean(cat(3,C{:,k}),3);
end
