You'll have too much variation in sites to get anything accurate with only two maps and two data points per map. Z-levels, weather, creatures, idle dwarf behavior, finicky processor, alignment of the planets, etc.
Bar an accurate sample, you can just embark on a good number of sites and note the statistical measures of the FPS for each: mean and standard deviation are typical, mean is appropriate for something this coarse.
What you're calculating is an average. Since there's no proper statistical measure for the average between minimum and maximum values, it's just a plain ol' average. Since it's very possible that you captured outliers, then it's also very possible those aren't very accurate measures at all (please don't take offense; it's not wrong, just not complete). A better method is to take many FPS readings for each map. Just sit and watch the FPS counter and note down every number your eyes register (a precision of 10 FPS is perfectly fine). After you've written down a good number of them, take the average of them all. That's a fair measure of FPS for each map.
To relate this to map size, then embark on a number of NxM maps with a lot of variation: water, magma, HFS, chasms, aquifers, z-levels, biomes, civs, and pets just for the hell of it. With that, you have an average for the FPS as related specifically to map size (by the way, you can create a distribution for each of these measures to get a better idea of how the numbers skew).
Finally, to create a relation between map sizes to FPS, you need more than 2 map sizes. Two can only give you a linear relationship at best. Three is necessary for anything non-linear. Five or more is good to verify your model (like ed boy's). To simplify things, you can discard multivariate map sizes (NxM) in favor of map area, but that's a critical assumption.