All progression resources (production, gold, science, culture) and the rates that you earn them at, all scale proportionally with game speed. If something took 100 production to build in one mode and 200 in another, then it takes twice as many turns, so the game speed doesn't affect the difficulty with respect to any of the progression resources. Using one city with 5 production as an example. In the faster game, it would take 20 turns to build it, and 40 turns in the slower game. The tempo of the game actually stays constant across all the game speed settings when it comes to those 4 main progression resources.
But once a unit is built, how it's used differs drastically with respect to its build time. The main thing that makes AI weaker on marathon is unit micro management. You can perform movements and attacks with it each turn. How efficiently you choose to move and attack with your units can gain you marginal gains with each decision and action you make. For example, you might find more goodie huts, or you might rack up more xp and gold clearing barb camps, and most of all, you might conquer more cities than an opponent with the same number of turns to perform unit actions.
Marathon allows you to maximize this unit micromanagement advantage over the AI because it spends most of its turns shuffling its units in place, or chasing off barb scouts. On the other hand you move your units every single turn with a sense of purpose, inherently more efficient.
The second micromanagement advantage (less significant than unit actions, but still significant) is production rounding effect. When you're building a unit naturally, the production cost overflow is generally wasted. If something took 200 production to build and your city now has 5.5 production instead. After 36 turns you'd have 198/200 complete. The final turn gets another 5.5 production even though you only needed 2, so only 3.5 hammers are effectively wasted. You might adjust your tiles around in the final turns of production to get a free point of science or gold or whatever from different tile working choices.
In the faster game it would take 100 production to build so you'd finish it in 19 turns instead (99/100 on turn 18, and the final turn wastes 4.5 hammers). Even assuming perfect tile swapping so you don't waste anything, this 37 turns vs 19 turns is already not 2:1. The slower game already gained a 5% edge (37 to 19 is 1.947 to 1). Early on, this difference is more significant, and they really add up and snowball harder over future turns. As the production numbers grow larger, the edge grows even more, although it becomes less significant. At 45 production the fast game would take 3 turns to build, and the slow game would take 5, again not 2:1, but here the slow game gains an edge of 40% per instance (1.6 to 1). When you slow down the game speed, this "rounding" effect either makes no difference or is in favor of the slower game. It'll never be in favor of the faster game
And of course all these marginal advantages compound geometrically, so in 200 turns of a faster game vs the equivalent 400 turns of a slower game... if the above example is some kind of military unit. Not only do you end up with 66 units in the faster game vs 80 of the same units in the slower game, each one of those units have been moving and attacking more efficiently throughout the game. You'll simply end up with more stuff, alot more stuff.
