Killer apps, when it comes to languages, are feedback loops, but you need a statistically significant number of people using those languages to make it more likely that those killer apps get made. To me, this really just sounds like statistically no one used Self so no one (or not many, to be kind) made useful apps in Self.

imo, the Self/Smalltalk vision was just hard to sell to people in the 80s/90s since it was so ahead of its time, with no easy bridge back to the world people were building their existing software in. It might even be hard to sell it today. People built their apps with an existing toolchain and a language they already knew. To adopt the Self/Smalltalk vision largely meant ditching all that to replace your language, your OS, and your toolchain at the same time, for a bet on the claim that a dynamic language was somehow fast enough.

Instead, we piecemeal evolved into a worse world where Javascript apps run in a worse OS (the browser) and Python apps run in a lower level OS that needs to transport itself in a worse image model (Docker). But, it was easier to adopt from what came before. I can't say I'm happy about it, but at least the vision of Self and Smalltalk caught on.