Look up the Barna statistics. It shows you divorce by denomination and church attendance as compared to the world. IIRC in that study the failure rate of any given marriage in the first 10 years is around 42% or so for non-church goers. Differing studies and stats in different areas come in at different rates and so overall it colloquially came to be thought of as 50%; which isn't far off really, not in the grand scheme of things when is ought to be an order of magnitude less.
The Barna rate for the world is also not far off the church average of around 36 or 38. (10% diff, not far off the margin of error). And Protestants and Catholics differ little on that.
I don't know where he gets the NFP rate, but it doesn't surprise me. That's basically selecting for staunch Catholics; who worldwide have very little divorce (except in the USA, dominated by our cafeteria Catholics and lax diocese). The US Catholic church grants more divorces than all other countries combined. IIRC an order of magnitude more even.
NFP doesn't just select for Catholics who will be willing to follow church tradition in other ways (i.e. divorce); it also marks people willing to set aside the world's standard for marriage for the churches.
Praying together, now you're dealing with a wife willing to bear her soul before her husband. That's quite out of the ordinary and creates a different dynamic.
Most current Barna survey, which differs a little from my numbers. The 42% number above must have been from an earlier survey where they listed by church attendance (couldn't find it just now), or from US wide stats. Again, off my memory. But it does show all adults at 33% and Protestants and Catholics at or a little below that. My recollection matches O'Brian's and is from an earlier survey.
A note about Barna. It is based on survey data. Doing it that way allows him to break it down by church affiliation, something you can't get from county marriage stats, but doesn't capture those who have left the church; which is why I think the rate fell from 38% to under 30 in the newest survey. It also exposes you to further error due to people not honestly self reporting.