this can be refined to the point of absurdity in a lower difficulty game.
Consider a monarch level game on a huge map with 31 opponents for example. the player is the tech leader and is at war with his neighbor and about to eliminate him. Player has 1000gpt currently.
1) micromanage all cities to maximize commerce, NOT food. gpt --> 1150 gpt.
2) sell your only iron, horses, etc for more gpt provided there exists an ai who has spares to sell to you. gpt -->1400 gpt.
3) give away all gpt for alliance to ONE backward ai. sell much tech to same ai for all gpt back.
4) now do this to second ai.
5) repeat process 31 times starting with most backward ai to most advanced netting 20000+ gpt.
6) buy back iron, horses, etc. re-micromanage cities, etc.
this is such an enticing exploit i would be tempted to actually install civ III and start playing it again ... but ... no ...
edit> hmmm, thinking about it more, to get the most gpt possible it is necessary to research as fast as possible. so i'm thinking to make this even better the player should irrigate everything possible as if it were a milking game and turn his entire core into a scientist farm. as for the big gpt he should rush libraries and unis everywhere. this will give him rapid science progress so that he can be ready in 20 turns with 5 new techs to broker around again.
edit again> in a huge map deity game with 31 opponents there will be many civs that are not behind in tech. no matter ... there are enough that ARE behind that this technique could be used to catch up and race ahead. .... makes me want to play a new sid level histographic milking game with 31 civs ...
ok one more edit> in this hypothetical 31-civ game no one will be able to trade for many luxuries until quite late. so now if i'm raking in huge gpt then this strategy would actually dictate that i rush happiness improvements everywhere allowing me to put my science rate at 100 percent.