Scientifically, our general rule of thumb is to maximize statistical power. That means run as many independent samples as possible. Once the number of independent samples is maximized, it then makes sense to begin to replicate. How many replicates will be is hard to say. Every case is unique.