So far I've been using a Zen-Fire demo account for access to historical minute data for backtesting. I believe this data originates from Rhythmic. I'll probably try another demo account with another provider which I believe is also based on Rhythmic.
But to answer your question... I've been using a 1 minute time frame data series for executions for better backtesting. The resulting equity curves are less smooth and more likely more realistic. So for example, I'll try one time frame for entry (say 60M) and another time frame for exit (say 30M) and each will execute on the 1M time frame.
You might also stay away from the Genetic Optimizer if you're getting greatly different results compared to the Default Optimizer.