A pitch of a given size looks larger on the map (at the same zoom level) the further it is away from the equator; two examples with perimeter ~ 110 m:
* https://www.openstreetmap.org/way/1029667455 Quito, Ecuador, lat ~ 0°
* https://www.openstreetmap.org/way/88871677 Northern Norway, lat ~ 70°N

But regardless of this, length() can still give the correct length in metres. Here is an explanation how it works. If it works correctly, I don’t understand why “it is only possible to do this sort of comparison for relatively small areas.”

osm.wiki/Overpass_API/Overpass_QL#Length says: “The length operator returns the length of the element in meters. For ways this is the length of the way.”

Are you saying that this is not correct, that I need to somehow convert the value returned by length() to meters depending on the latitude?

Overpass cannot calculate areas, but it can calculate the length of a closed way. I looked at a few dozen tennis pitches with different outline lengths and got the impression that almost all (>95%) pitches with an outline length < 115m and at least 80% of pitches with an outline length between 115m and 120m contain only one court. Check it out for yourself: https://overpass-turbo.eu/s/1uEl

Worldwide we have 347525 pitches mapped as closed ways with length < 115m, 10440 with length between 115m and 120m and 443169 in total. It follows that at least (95% * 347525 + 80% * 10440) / 443169 = 76% of all tennis pitches mapped as closed ways contain just one court. Obviously a much different result. Where does this big difference come from?