What I meant to say was the less correlated the resulting equity curves are the better the systems hold up against drawdowns. But the equity curves must have positive expectancy or you are just playing in sinking sands. Well you backtest a system on a time series such EURUSD and zorro should produce an equity curve. If the equity curve does show growth and you want to add another profitable pair to the system, you run the same logic on the new pair and build a mean-variance matrix on the equity curves and do the portfolio optimizations as the finance text books preach.
Here is a link link that discuss the topic in-depth.
But I personally look to use a 1/3 portion of the optimal f in allocating funds to trading systems in my portfolios.