I would use 500 bars minimum. Unless there has been an obvious structural change in that market during that period.
If you haven't read Bruce Babcock's "The Dow Jones-Irwin Guide to Trading Systems", you might find it helpful. Robert Pardo's "Design, Testing and Optimization of Trading Systems" is good also, especially because it's still in print.
With any kind of luck, these postings will bring old Jack out of retirement to rail against the backtesters.
Probably Jack would be an expert in backtesting and automated trading, according to some of his posts I read.
Jack's telling people that his methods cannot be backtesed and backtesting is not useful would be a good way to drive people to pay more attention and follow him without too many questions. Perhaps he uses backtesting at home frequently.
I think you also might want to ask yourself the question after how many days (some number between one and never) of trading you're planning to perform another set of backtests, and perhaps calibrate. Then you might want to do a series of walk-forward-tests using all your data (e.g. 4000 bars) and different values for your optimization period X (e.g. 400 bars), and your (hypothetical) trading period Y (e.g. 40 bars), see if any patterns emerge and look for more-or-less robust combinations of X and Y, judged by a variety of statistics of interest (e.g. annualized return, dd, sharpe etc.), estimated over the (hypothetical) trading periods (not over the optimization periods!). This still doesn't guarantee anything for the future, but it's probably more informative than your current approach. E.g. if your system shows acceptable outcomes for a broad range of X and Y values, you might have something valuable. If it's very sensitive to X and Y, you might want to try something else.