As you correctly pointed out, it's all a matter of variance. This is like experimenting with a new treatment on patients. First you give them treatment A (in your case, the 3.99 treatment). Then you give them a new treatment. The 2.99 treatment. And you measure the results. You know what your income was with treatment A (for this, take the average daily income over a month). That's your reference level. Let's say you set the variance at 10%. Everyday you note your income (price*sales) with the new treatment. Statistically speaking , if the point is within the variance levels, it means that the point is not significant. Or in other words, you cannot conclude that the new treatment is better or worse than the previous treatment. And this can go on forever. But in practice, for every point under the reference level , you're losing money. So you might decide to stop it after a couple of days.

Every point that is outside the variance levels is statistically significant. Which means that you can use these to draw reliable conclusions. Now how should you set your variance level? That's a bit more complicated, because that level depends on the price of your book and the conversion rate. (How many viewers will actually buy). You can do this with a chi-suare test. But I wont bother you with annoying statistics stuff. For a 3.99 book, the variance is, depending on the conversion rate

conversion rate->variance%

6%->13%

5%->14%

4%->16%

3%->18%

2%->22%

Another more simplistic way to do this, is to draw one line at your reference level. Then take all the points that are above the line and calculate their average. That will be your upper variance level. The same for the lower variance level.

But if your selling thousands of books a day I think you can throw in $500 for this experiment

Let us know how it goes