If I remember rightly, former senior Boost folk tried to stop std::variant shipping with its valueless-by-exception design, demonstrating that it could be easily avoided using a fallback to a double buffer design.
Peter and I were both very vocal in committee lists about this, although I was not too fond of double-buffering specifically. There were several multi-hundred post threads leading up to C++17 about it.
As I mentioned, the outcome could have been worse still. And there are much worse mistakes-in-hindsight in the standard. Though, I guess we are actually finally getting C++ 98 exported templates - albeit somewhat reworked - in C++ 20. I look forward to seeing if that damning EDG paper on exported template link performance gets borne out in real world applications by average programmers who don't fully understand Modules. Niall