From my observations: cold start, ease of patching.
If you're running a lot of different JS code or restarting the code frequently, it's faster than node.
Where it's useful: fuzzing. If you have a library/codebase you want to fuzz, you need to restart the code from a snapshot, and other engines seem to do it slower.
It's also really easy to patch the code, because of the codebase size. If you need to trace/observe some behavior, just do it.
Because the main goals are to be small, embeddable, and have low startup time.
A top class engine requires a JIT and that generally leads to being large and having high startup time. LuaJIT is pretty much the exception which proves the rule and 1. Mike Pall is a monster of a developer not unlike Bellard; and 2. Pall is on record (in discussions with other JIT people) that in his opinion you can’t bolt a JIT to an existing runtime and get a maximally efficient JIT. So you’d need someone of that class of skill to decide to make that their legacy.
And it's not like that Bellard is unable to produce a JIT compiler; see, for example, QEMU dyngen [1] which eventually inspired CPython's new JIT compiler.
Like other replies said, having JIT requires quite different tradeoffs to be made.
And unfortunately, making a very fast interpreter-based JS engine is a tedious and human-power-heavy work, it involves a lot of work which can't be eliminated by better skills.
Apparently Figma still uses it for plugins, so there's that.
It can be compiled to WebAssembly and run with the sandboxing of WebAssembly. Not just data sandboxing but sandboxing that can prevent it from slowing down the rest of the app (though it runs slower).
This in 2023: "Either Figma hasn't updated QuickJS since, or they've moved to a different VM, because Figma's VM is missing many of the features QuickJS now supports." https://www.sam.today/blog/bumbling-the-figma-api
