This is a normal statistical relationship. But it is all really an artifact because you still have to fill up the tank every time it is empty. All the variability that is introduced at that time is still there whether you record the MPG then or not. In order to get more precise MPG data you would have to actually double the size of your tank, not just skip a data point between two tanks. It is best to take the measurements for every tank because you then have the most versatility with your data.

How many tanks you should average in your spreadsheet depends on what you want from your data. If you want the best absolute estimation of your car's mileage, average all your data points and have one average MPG figure (but that's a very boring chart

). If you are interested in seeing how your mileage changes over time in response to changes in driving habit, season, or aero devices you should pick shorter measurement periods (per tank for example) and put up with somewhat higher variability from the true average. If you are interested in how your mileage is trending just have your spreadsheet calculate a trend line using individual tank measurements.