Remember that ANY connection that is not welded (soldered) is a loss of 0.75dBm to 1dBm. As you know, a 3dBm increase in power means the Tx power is doubled, as is any decrease of -3dBm the Tx power is cut in half.
I do not put a lot of faith is testing the Tx power through a solid connection (connecting the Rx meter to the RC by cable) since what you see is not the true RF power received by the drone. Also you can have the best and most ideal Tx RC reading BUT if the Rx in the drone is weak, all the calculations are off the table. Not to mention, is the P3/P4 equipped with an LNA (low noise amplifier) to filter out all the noise and unwanted RF in the area? What about the RC receiver (Rx) does it have an LNA? To improve Tx and Rx of any units, Rx diversity is needed and the Div Imb should not exceed 3dBm (diversity Rx is achieved by having 2 antennas capable of RF Rx and should be calibrated within 100mHz band separation). In DJI case, RC and drone have ONLY 1 Tx and 1Rx antenna which makes it a low ball RF unit.
The kind of testing we can do here - absent a formal test range and an anechoic chamber - is fairly primitive and
always relative and valid only under similar conditions. I have been putting some thought into this, trying to figure out exactly how to test different antenna designs and it's frustrating. And I have lots more test gear than the average UAV owner.
The likely best you can do is
1. Find an area with low background 2.4 GHz noise.
2. Use the same experimental set up all of the time.
3. Minimize use of connectors, keep cables short and symmetric
4. Do a number of tests on different days and keep good records.
5. For actual flight tests you will need low noise conditions throughout the entire test range.
6. For actual flight characteristics you will need to check ranges at different vertical and horizontal angles from the RC (beam width considerations).
7. You should also keep relative humidity fairly constant for distance tests.
So, by just stuffing a power meter on an RC you will get some broadly entertaining bits of information, but of limited utility. You really don't give the northbound end of a southbound rat about the numbers, you want to know if this configuration is better than another and under what circumstances.
The numbers I see here fall into the 'interesting, but what quite does it mean' category. About the only data that is broadly applicable are the folks looking for maximum range who run up and down the same course all of the time. Even there, subtle changes like using different RCs are probably not going to be experimentally verifiable. For a more generalized use case, beam pattern characteristics are really important. And for the general public, even factors such as size, configuration time and robustness may have more applicability than raw power.
So keep at it folks, just be careful not to get over comfortable with your data and realize that if you are actually trying to make some sense out of all this, it's pretty complicated. At the very least, doing these sorts of experiments can get you an appreciation of how hard GHz radio transmission is and how much performance DJI has managed to stuff into a cheap, small package. Yes, DJI could do a lot better and I expect the next generation of transmitter receivers will have some improvements in bandwidth rather than pure power as we're already at distance numbers that give the FAA the heebiejeebies.