wouldn't it be simpler to buy 2*Gold than Capital+1*Crypt?

I had that exact same thought. Actually, it's Capital + 2*Crypt, because you need two crypts for each Capital if you're setting it aside every turn and drawing deck.

Wait, let's do the math.

6 Golds makes $18 for a double-Province and change. So you technically only need one additional card worth of draw (and a +buy. So one Margrave overdraws your deck by one card.)

3 Capitals makes $18 and gives you the buy. It would take 6 Crypts to set them aside. However, every turn STARTS with six cards on the table (three Capitals and three Crypts) and only 3 cards in the rest of your deck! Turn is play $18, buy Provinces, play three Crypts, set aside three Capitals, clean up the 3 Crypts from the previous turn.

It needs NO draw, and can even work with 100% reliability after the first double Province buy! The degree to which leaving cards on the table between turns helps a deck-drawing engine keeps surprising me, even though it shouldn't. Nod to Gear and Archive.

So the math says: 9 cards at $5 each is 45, 6 cards at $6 each is 36, plus Margrave or equivalent puts you at a similar cost, but with less reliability. The math looks good on this one. With decent draw it could be huge, though the Crypt split seems pretty important.

Now, the math on keeping Cursed Gold around just because you can reliably trash the curse every turn in your deck drawing engine, on the other hand... (yes, I saw someone do this. I pointed out that between the Cursed Gold, the Curse, and the trasher, they were using three cards to generate $3. That's Copper the hard way.)