Why Eigen is 5x slower than ublas on following example?

At Eigen version I use "true" fixed size matrices and vectors, better algorithm (LDLT versus LU at uBlas), it uses SIMD instructions internally. So, why it is slower than uBlas on following example?

I am sure, I am doing something wrong - Eigen MUST be faster, or at least comparable.

#include <boost/numeric/ublas/matrix.hpp>
#include <boost/numeric/ublas/vector.hpp>
#include <boost/numeric/ublas/lu.hpp>
#include <boost/numeric/ublas/symmetric.hpp>
#include <boost/progress.hpp>
#include <Eigen/Dense>
#include <iostream>

using namespace boost;
using namespace std;
const int n=9;
const int total=100000;

void test_ublas()
{
    using namespace boost::numeric::ublas;
    cout << "Boost.ublas ";
    double r=1.0;
    {
        boost::progress_timer t;
        for(int j=0;j!=total;++j)
        {
            //symmetric_matrix< double,lower,row_major,bounded_array<double,(1+n)*n/2> > A(n,n);
            matrix<double,row_major,bounded_array<double,n*n> > A(n,n);
            permutation_matrix< unsigned char,bounded_array<unsigned char,n> > P(n);
            bounded_vector<double,n> v;
            for(int i=0;i!=n;++i)
                for(int k=0;k!=n;++k)
                    A(i,k)=0.0;
            for(int i=0;i!=n;++i)
            {
                A(i,i)=1.0+i;
                v[i]=i;
            }
            lu_factorize(A,P);
            lu_substitute(A,P,v);
            r+=inner_prod(v,v);
        }
    }
    cout << r << endl;
}

void test_eigen()
{
    using namespace Eigen;
    cout << "Eigen ";
    double r=1.0;
    {
        boost::progress_timer t;
        for(int j=0;j!=total;++j)
        {
            Matrix<double,n,n> A;
            Matrix<double,n,1> b;
            for(int i=0;i!=n;++i)
            {
                A(i,i)=1.0+i;
                b[i]=i;
            }
            Matrix<doubl