I'm a bit torn. Quite obviously, exploiting the Taj Mahal is a lot more powerful, especially if you have Chichen Itza and play with Persia. The thing is that Golden Ages are more powerful for large civilizations but to amass a useful amount of excess happiness you need to invest large amounts of money or have only a small empire.
All in all, Golden Ages are good for hammers but bad for science. In my opinion, for large empires the excess amount of stuff you get from the golden age does not, in itself, warrant the cost for creating happiness buildings that you would need to obtain a happy golden age, while for small empires the benefits of the golden age are marginal.
You might also want to take into account that you can sell luxuries to the AI for a cost of 5 happy faces for an equivalent up-front payment of 10 gpt. That is 2 gpt per happiness point. So even if you have "free" excess happiness due to luxury resources, selling them to the AI and investing the money might be a better choice.
If you like, here's a bit of mathematics:
Let's say we try for the second happiness GA. This costs 1000 happy faces. Let's say we have 25 happy faces, this will mean it takes 40 turns to get the GA.
If we assume we could have the 25 citizens work tiles during those 20 turns, they should average something like 1.5g + 0.5h + 1s each. So we get 25*20 = 500 yield turns which means with my assumption of average yield 750 gold, 250 hammers and 500 science.
During the golden age, each tile yields an additional +1g +1h Happiness golden ages take 10 turns so for each citizen working tiles in your empire you get +10g +10h extra due to the golden age.
Comparing this with the above we can get the number of tiles you need for the GA to make sense at the expense of growth. If we assume 1h = 1g = 1s for simplicity, you need 75 tiles worked to break even. Obviously, if you are Persia or have CI, you only need 50 tiles and if you have both you only need 38 tiles.
This analysis becomes obviously worse for the later golden ages. So from this I would say that probably only the first golden age really makes sense. After that, I would rely on great people and the Taj exploit.