We performed numerical simulations of supersonic isothermal turbulence driven by mostly compressive large-scale forcing, using both a static grid and adaptive mesh refinement with an effective resolution N=768^3. After a transient phase dominated by shocks, turbulence evolves into a steady state with an RMS Mach number about 2.5, in which cloud-like structures of over-dense gas are surrounded by highly rarefied gas. The index of the turbulence energy spectrum function beta = 2.0 in the shock-dominated phase. As the flow approaches statistical equilibrium, the spectrum flattens, with beta = 1.9. For the scaling exponent of the root mean square velocity fluctuation, we obtain gamma = 0.43 from the velocity structure functions of second order. These results are well within the range of observed scaling properties for the velocity dispersion in molecular clouds. Calculating structure functions of order p=1,...,5, we find for all scaling exponents significant deviations from the Kolmogorov-Burgers model proposed by Boldyrev. Our results are very well described by a general log-Poisson model with a higher degree of intermittency, which implies an influence of the forcing on the scaling properties. Contrary to previous numerical results for isothermal turbulence, we obtain a skewed probability density function of the mass density fluctuations that is not consistent with log-normal statistics and entails a substantially higher fraction of mass in the density peaks than implied by the Padoan-Nordlund relation between the variance of the density fluctuations and the Mach number. In conclusion, it seems necessary to account for the production mechanism of turbulence in the ISM.