Caravan revenue is based on the trade in the two cities times the distance separating them...
One has Colossus/Copernicus with a GemSwamp and plenty of fishing squares, and every land square on the planet is railroaded (except the Poles. I can land a diplomat on the South Pole, but not a Settler. Pity.) That city has 65 trade-arrows and lights up 218 bulbs per turn. The second-highest trading city has Gold and a nice river vale, and it's also 65 trade-arrows, on a separate continent. It racks up 89 bulbs per turn (i'm at 20Tax-70Sci-10Lux).
... you are getting 2/9 of the value you would be getting from a strong AI city. 300gp * 2/9 works out to about the 67 you are observing.
Alas, there will never be an AI city, since the last remaining unit is a Zulu Settler checkmated onto a mountaintop. When that happened, got goosebumps and decided to try and exploit the opportunity, to break the game in unusual ways. Your info gives me another interesting idea:
Routinely do Colossus/Copernicus in the same city, and sometimes i'll put Shakespeare there too, if the scenario makes it worthwhile. And in previous games, i have used the Diplomat Sneak to run a Settler into enemy territory and railroad their grasslands for them, just so i get better trade. What if...
What if i choose an enemy city exactly half the world away from my Col/Cop city, and refuse to kill that one city? I can set them up with massive trade whether they like it or not, avoid the same-civ and same-continent penalties, and if that's the civ's only remaining city, then they'll never get RR or Flight (assuming i stunt their expansion by 1000 BC).
Not just a pure theoretical experiment, if this is possible then it means you can overcome a tax deficit of 1000 gold pieces per turn, just by buying five Camels for less than 98 gp each. Woo, that kind of income can support a perpetual parade with Luxury at 80%. My side-game has a Plains city with access to three Golden Mountains, so if i can find a suitable victim-city 1/2 the map away, might be fun to try out this idea.
You can set the research rate to 0% and still make discoveries from caravan trade. If you are trying to avoid the pollution trigger advance, you have to stop trading when you get to FT 187.
This is very useful to know, thanks. A correct strategy is taking shape. Getting 3 FutureTechs per turn, so perhaps: restart back a few years to 175 FT's. This should allow me to perfect the trade network without breaking 188 FT's (i need about 30 more Caravans to get all 117 cities paired up with maximum trade-partners). Then drop Science to 0%, bring Tax up to the lowest positive number, and jam the rest of the trade-arrows into Luxuries. Then it's Beethoven's Ninth all the way to 2030, when i start the SS. 2041 it launches full size, and in 2059 it lands.
In 2054 rebalance trade to get up to 190 Future techs by 2057. By now, the pop-effect of parades should have maxxed every city with a wheatstalk to spare, running hi-lux from 2015 to 2054, so i should be able to afford the Science. Finally, in 2057, jump up to 100% Luxury. Full Peace bonus, full SS bonus, full Wonders and +950 in Future Techs.
By 2011, had civscore 3926 (Prince, 7 civs). Wondering if i can still push this game over five grand. Would have made 5,000 easily, if i could have kept up the FT's past 188. Was on a pace for 330 FT's by 2059 (+1650 civscore).
Alas, pollution.