Correct. What it does is prevent data races; this is per se nothing new, but was done as early as the 70s [1]. There've been various and sundry approaches to the same problem over the years (the 90s and early aughts produced a lot of research in this area).
In a way, it is frustrating that so few programming languages offer prevention of data races, so I'm grateful that Rust pushes that; but in another way, it is understandable, because all such mechanisms restrict what you can do, often in undesirable ways. Google "benign data races", for example: data races can be purposely exploited for more efficient implementations.
The other problem is that data races are really the easy part to solve about concurrency. The hard parts are general race conditions, non-determinism/causality, liveness properties (such as freedom from deadlocks and starvation), and performance.
Performance is one of the trickier aspects: in a shared memory system, performance is primarily a matter of reducing contention. But virtually any system that relies on some form of mutual exclusion to prevent data races introduces contention, and you have a constant tug of war between the two concerns.
Particularly frustrating is that all of these issues are what software engineers call "cross-cutting concerns" [2], i.e. things that cannot easily be modularized. For example, performance may suffer from a serialization bottleneck in a module's implementation where you cannot avoid that bottleneck without exposing the module's internals.
