I would not be too worried about the waveform above and below the crossing point, as long as the zero crossing is equally spaced and consistent with load it should not matter. I would be more concerned about noise spikes giving you a false crossing.

Well, I already have that problem, or a kind of: every half cycle the ZC signal is not crossing zero volts.
Better look this scope capture:

 

Ok here are my 2 cents:

This is my all-time favorite Zero-Crossing-Detector circuit:

It dissipates very little power, it's dirt cheap, it's super small and uses low-voltage, garden-variety parts.

(This entire thread is a lesson - tell people what you really are trying to accomplish- rather than the tail wagging the dog)

 

Ok here are my 2 cents:

This is my all-time favorite Zero-Crossing-Detector circuit:

It dissipates very little power, it's dirt cheap, it's super small and uses low-voltage, garden-variety parts.

I also tried that circuit, or a very similar one. I found it super interesting for the same reasons that you give, but I discarded it because I didn't get an explanation about its working (I'm far from being an expert in electronics) and how to choose components values in order to tune the ZC signal closer (or farther) to the real zero. That is: I "need" to understand what I'm doing and why it works.
Nonetheless, the document you attached has a full explanation about this solution. I think it is worth a new revision.
Very grateful for your contribution.

 

Well, manufacturer's datasheet states that under no load the transformer draws 0.15W at most, that's a lot less.

I also said if you didn't need the characteristics I required you could do it in a lot less, try reading peoples replies to you!

 

Ok here are my 2 cents:

This is my all-time favorite Zero-Crossing-Detector circuit:

It dissipates very little power, it's dirt cheap, it's super small and uses low-voltage, garden-variety parts.

And it's phase-accurate, unlike a transformer which has a phase shift due to Lmag and Rpri. With 440k in series with the mains, it only dissipates 120mW, which is less than the transformer.
If you need one that is simpler, then a resistor and a PC814 does the trick, but it needs more current, pulse width varies with voltage and it is not guaranteed to be symmetric (because the two LEDs in the PC814 are not quite the same, nor in the same place relative to the phototransistor)
But if you are just counting cycles, not triggering triacs, it's good enough.

 

For measuring, I might also mention the ZMPT107-1 transformer available from quite a few sources.
http://www.interplus-industry.fr/index.php?option=com_content&view=article&id=53&Itemid=174&lang=en

 

For measuring, I might also mention the ZMPT107-1 transformer available from quite a few sources.
http://www.interplus-industry.fr/index.php?option=com_content&view=article&id=53&Itemid=174&lang=en

How thin is the wire that they wound that with?
I've also seen some Hall effect devices for the same purpose.
https://www.lem.com/en/product-list?measurement=82&nominal_val=200-400
At the price they charge, there must be good reasons to use one instead of a transformer, although I can't think of one.

 

How thin is the wire that they wound that with?

It doesn't need to be thick to handle the specified 2mA signal. This is a measurement transformer. It is not intended as a power transformer.

 

It doesn't need to be thick to handle the specified 2mA signal. This is a measurement transformer. It is not intended as a power transformer.

It obviously has a very small core, with a very small cross sectional area, so that it needs a huge number of turns. The number of turns is no different whether it is a signal transformer or a power transformer.