As you found, the solution is in termination. The problem with series termination is that it should be near the transmitter, but on a bi-directional signal the driver changes. This is why DDR3 uses On-Die Termination (ODT) that can be enabled and disabled dynamically. From the description of your results, it appears that the ODT in your controller is not available, or your simulation setup does not enable the ODT. Look into the controller IBIS model to see if it includes the option for ODT and then verify that your DDR3 wizard setup uses the correct models for ODT enabled configurations.