This from here seems pretty accurate for Usenet:

Binary groups being a great big cost sink would be the main thing.

The store and forward protocol required quite some disk space at the time.

The network relied on "control" messages to create/delete groups automatically (as opposed to manual subscription), which due to the lack of authentication/encryption in the protocol, were very easy to spoof. A gpg-signing mechanism was later put into place, so that nodes peering with each other could establish a chain of trust by themselves. This was pretty nice in retrospect (and awesome by today standards), but the main problem is that creating new groups was a slow and painful approval-based process: people often wanted small groups just for themselves, and mailing lists offered the "same" without any approval required.

Having a large open network started to become a big attractor for SPAM, and managing SPAM in a P2P network without authentication is a harder problem to solve than a locally managed mailing list.

running a local server became so easy and cheap, that running mailing list offered local control and almost zero overhead. People that had niche groups started to create mailing lists with open access, and people migrated in flock. Why share your discussions in comp.programming.functional where you could create a mailing list just for your new fancy language? (it's pretty sad, because I loved the breadth of the discussions). Discussions on general groups became less frequent as most of the interesting ones were on dedicated mailing lists. The trend worsened significantly as forums started to appear, which lowered the barrier to entry to people that didn't know how to use a mail client properly.

For NNTP for LessWrong, I would think that we have to also take into account that people want to control how their content is displayed/styled. Their own separate blogs easily allow this.