That road (C or Rust parsing core through bindings) will likely be taken, but for the initial development and jump-starting the ecosystem it was important for me to start with implementations that can be quickly experimented with and iterated and not spend a lot of extra time on dealing with segfaults, memory leaks, the different binding mechanisms on different platforms, etc. As things stand now, people are provided with multiple, fully functioning, pure implementations that already are faster than the majority of YAML/TOML parsers. In the coming months and years there will be plenty of time to make things even faster. :)
For me caring about performance also means caring about performance on all platforms, why not after all? You can take the tabular benchmark data I provide and paste it together, or use the raw data that is also available as eno files in the repository to compare language against language too (which I initially also did but later dropped because same-language comparison for libraries made more sense to me), if you want the quick run down as far as I remember it: mostly javascript parsers lead the ranking, ruby parsers are a bit behind and just slightly ahead of Python.