Because it is "just Ocaml", all the weird parts of Ocaml bleed through the reason language design. No operator overloading, nominal record typing, 3 competing standard libraries (plus the all-too-small actual standard library). The list of oddities is rather long.
SML was and is a better choice. SML has structurally typed objects and first-class constructors. No named arguments, but structurally typed objects do a better job at this (like JS destructuring, but more efficient). Anonymous records are possible. A ref type rather than mutable record fields. Only one `let`. PolyML has shipped a multi-process implementation for years while Ocaml's is still in an unending beta. SuccessorML (SML 2.0) is adding some syntactic niceties along with things like module typeclasses (modules should prevent the typeclass hell of Haskell). Most importantly, SML is a standard with multiple implementations and isn't completely tied to the success of Jane Street.
Our university was big in Prolog (SWI) and Caml Light back in the day, and they still are, just with Caml Light having been replaced by OCaml.
The only reason I known a bit about SML is because I am a language geek, it is hardly known in this side of the pound.
That's why I first loved it. I liked what I knew of Haskell but was skeptical of the Monadic IO system (I've since warmed to the idea). OCaml was in a perfect sweet spot in my opinion.
Also it feels a bit less abstract in some difficult to define way - I don't think Monads really clicked for me until I was using Options in OCaml.
I'll also second the opinion that as a _language_ OCaml is more ergonomic than Rust. However, I find that Rust is more _community ergonomic_ - it's easier to find libraries and contribute to them. Related to that - the build process is just spot-on. With OCaml, I found myself tripping over obscure errors in my build process about every month or so.
For me, OCaml is a bit like the one that got away.
But I'm excited about the momentum behind the Reason project!