The biggest win, in my experience, is simply recognizing when you don’t need all ~16 significant digits provided by the standard implementations. If you’re only printing e.g. six digits, and can afford to round in the wrong direction 0.00000001% of the time, you can skip a ton of work. And as you mention, it gets even easier when you don’t have to support scientific notation.

The standard library functions occupy an awkward middle ground between this type of much faster, slightly sloppy float-to-string converter, and direct storage of the binary representation; in most applications where float serialization performance is actually relevant, at least one of these two alternatives is better.