I think the comment that there is a relationship between FS2 levels and image detail level downloaded is right on the money. There's a clue in the TMC readme where it states level 14 is roughly 1 m pixel.
I have been doing a bit of testing to understand how the scenery works in FS2. I did this by simply using FS2 with scenery I had created and observing the results of the different methods I had used. I also needed to work out how the tiles work with each other to better understand why my scenery choices worked or didn't and where best to use more detail. Knowing this also helps understand why GeoConvert does what it does. For instance you thought you did everything right, downloaded the highest res imagery, but the scenery looks rubbish in FS2 or Geoconvert simply didn't produce tiles.
So I think that the download image quality, GeoConvert level used and targeting detail where it is needed are the holy trinity of scenery design for FS2.
A quick for instance. I had been using Bing maps in FSET and noticed that higher quality downloads sometimes weren't resulting in higher quality when viewed in FS2. I had already created level 14 scenery for a small airport not far from home using Bing map data downloaded at 0.5m pixel quality. So I downloaded the exact same area from Google but at .25m pixel as Google data seems to have higher detail (but pesky watermarks). I then used the same TMC to create my level 14 tiles from the Google imagery as used for Bing data.
I located an area that contained a house with a oval pool which was easy to find and small enough to show up any quality differences. To my surprise, the 0.25 m pixel Google data gave the same slightly pixelated view of the pool as the 0.5 m pixel Bing data. I then ran the Google data through GeoConvert at level 15. When I looked at the pool again the pixelation was gone and a much more detailed pool had appeared.
My conclusion is that downloading 0.25m pixel resolution data and then using on a level 14 tile is a total waste time. More bandwidth, more CPU resources, longer processing time were all used for no benefit on a level 14 tile. Perhaps AeroScenery could assist users from making this mistake either by preventing certain combinations or just a popup advising the combination used will not give any benefit. So I think there is some benefit in providing some protection in your program to prevent users from taking actions which would give no benefit or at least some advice. Perhaps the guide could be integrated into the AeroScenery Help button. It might also help reduce support questions and user frustration. eg the guide could advise that they should use 4 m pixel quality on levels 9 and 11 in large areas, but use .25 m pixel and level 14 in small areas around airport or areas of low VFR flying. In some areas you only are doing high level flying, so detailed imagery is wasted in FS2.
I am reasonably sure I have a handle on how the tile levels, tile sizes, and tile draw distance and viewing altitude all work together to create visually stunning scenery in FS2 with the least possible CPU and GPU load. I though I might document how I thought this all works for the more technically minded users out there as it does aid scenery design. If my assumptions are correct, then in my opinion the scenery rendering method use by IPACS is just plain smart. No wonder it makes FSX performance look like very old technology (which it is).
Creating the scenery that gives the best visual result for the least possible effort will also have a flow on effect in FS2, as that scenery will use the least possible GPU and CPU resources. This then leaves plenty of resource surplus available for allow users to run future developments such as ATC and AI aircraft. Also if you plan distributing scenery to other users some sort of standard will provide consistent results amongst users, minimise file sizes and maximise scenery quality. The last thing you want to happen is someone to download 0.149 m pixel quality to create level 9 to 15 scenery for the whole country, then blame AeroScenery for poor system performance.
Sorry its a bit long winded, as is often the case with my posts, but I wanted to cover the whole picture.