We managed to find the mistake. When defining the constraints the code was written as
@constraint(m, con[i = 1:n], sum(coeff[i,j].*x[j] for j = 1:n) <= bound[i] )
instead of
@constraint(m, con[i = 1:n], sum(coeff[i,j]*x[j] for j = 1:n) <= bound[i] )
which lead to a significant increase in model building time.
With this change the results are more reasonable and the array container is way faster:
