There's one (usually minor) caveat here, which is around strings. In .NET, like Windows*, strings are UTF-16. If you call into native code that uses UTF-8 strings extensively, then they have to be marshalled (when passed to native code) and unmarshalled (when received from native code), which takes additional time and memory. I've run into this lately, which has led me to write special CStr/CString types (idea borrowed from Rust) to avoid unnecessary, especially round-trip, (un-)marshalling.

* = Technically, Windows now supports the UTF-8 "code page" where applications can use UTF-8 natively, but all NT system calls use UTF-16, so the conversion still has to happen somewhere