If the data is anyway generated, you need not store it in t6 files. Just generate it on the fly in the initial run, and fill the price arrays with it.

That said, I know no way to generate synthetic data that could really improve backtests, except maybe in very special cases.