A decent volume on elementary mathematics is Mathematics for elementary teachers (Beckmann, 2010). The book is intended to help strengthen teachers knowledge of the mathematics behind the ideas in elementary curricula (especially reform curricula, I think). As such, it's often a good place to check for things like this.
Benchmarks (also called "landmarks") are introduced in the context of comparing fractions. When students are trying to determine which fraction is larger, $\frac{4}{9}$ or $\frac{3}{5}$, one strategy suggested is for students to reason about their relationship to some other number, like the fraction $\frac{1}{2}$ :
When we compared $\frac{4}{9}$ and $\frac{3}{5}$ by comparing both fractions with $\frac{1}{2}$, we used $\frac{1}{2}$ as a benchmark (or landmark). The fractions $\frac{1}{2}$, $\frac{1}{4}$, $\frac{3}{4}$, $\frac{1}{3}$, and $1$ are good to use as benchmarks. (p. 73)
It's clear from this text that the numbers are somewhat arbitrary; there's not meant to be a definitive list of benchmark numbers. Students would choose a fraction benchmark that helps them compare.
I can't say whether others use benchmarks the same way (a quick look at some other books I have within arms reach doesn't show up the term). However, the use here is clear: a benchmark number is a number useful in reasoning about a problem. In this case, the benchmark is used as a reference point for comparison of fractions.
The intent is to encourage reasoning rather than procedure. There are algorithms some students are taught to use for fraction comparison, which allow them to replace mathematical reasoning with a couple of memorized steps and some arithmetic. But reasoning allows them to practice conjecture, work through coming up with a justification for their answer, and eventually have a way to defend their answer other than "this is what the procedure produced."
I should think any useful number used in reasoning could be called a benchmark. For example, in my response to another question (seen here), I wrote about student reasoning that transforms a subtrahend into the number $2000$. In that case, $2000$ is useful.
Another type of mathematical reasoning that might benefit from a benchmark is estimation. Numbers can be replaced by nearby benchmarks that make for quicker calculation, if the aim is to just ballpark an answer (an often quite useful strategy for many real world applications).
In summary, I don't think there is support for a definitive list of benchmarks. The ones Dr. Beckmann provides are suggestions ("good to use") but the real test is whether they are useful to the thinker in the midst of their mathematical reasoning.
Works Cited:
Beckmann, S. (2010). Mathematics for elementary teachers. New York: Pearson Addison-Wesley.