No good support from the underlying processors, which meant that no one really used this, because even if you had support for this, you had a high performance cost.

That led to users wanting to find other methods to make this work.

And in general, the path has been either to avoid shared memory entirely (negating the need for tx memory) or using lock free algorithms (which won't worked with tx memory).

It is a great idea, in theory, but it proved impractical in practice.

The Clojure language makes use of software transactional memory.

https://clojure.org/reference/refs