By decreasing the mesh spacing by a factor of two, we see that the the number of iterations has increased by almost a factor of four. An increase is expected, since more iterations will be needed for information to 'traverse the domain,' as the mesh spacing is smaller. One may guess that we have been robbed. It took four times as many iterations on four times as many grid points, yielding a total of 16 times as many floating point operations by simply increasing the mesh resolution by a factor of 2. However, Jacobi's method employs a second order finite differencing scheme. This can be seen by computing the ratio of the infinity norms between the computed values and the analytic values for the two resolutions:
This is close to 4, which is what one would expect when using a second order method with double the resolution.