As a general rule, time steps should be as small as possible - of course, with respect to the final simulation time.
I think your second run with a maximum step of 1ms leads to incorrect results.
An indication for that are the "waves" you can observe in comparison with the first run.
More than that, to evaluate and justify simulation results one should have a rough feeling about the results to be obtained. Therefore, which result is more close to your expectations?