But Smalltalk doesn't have a dream. It's a language. So what dream did fail? Adele Goldbergs dream to build tools that help children to learn programming? That's a success, she is using them and kids are learning programming with Smalltalk based tools - Squeak carries some of those tools. Alan Kays dream to build the integration language of a graphical workstation? That was a success, the machine was built, it ran and worked. His dream of building a full blown tool for teaching programming and testing implementations in a extensive environment? Look at Squeak. A big success for the field it's built for - you can even try different VM implementations based on Smalltalk code. The dream of the Parkplace founders to get a development environment that spans several platforms and operating systems? Was a success, see the result in the current version of Cincom Smalltalk. Ok, it wasn't a success for the company itself, at least not in the long run - but then, most companies sooner or later drop out of business. The business model for Parkplace failed, but their tools are still working and are still used - is that failure for you?
Smalltalk _did_ change the world. It spawned whole languages like Objective-C that are nowadays used to implement an operating system that powers 5% of the personal computers in use. It influenced graphical environments more than most other system. It influenced development tools in many ways - hierarchical class browsers, the Python module path browser, many tools are modeled after what was already there in the original Smalltalk systems (and wasn't there before in this particular way - some Lisp machines did have similar tools, but they were still quite different from the much simpler view Smalltalk presented). So in what way did Smalltalk actually fail? There are not really that many languages and environments that changed so much in the world of software development as Smalltalk did. Generational garbage collection - first implemented in Lisp systems and Smalltalk systems. Virtual machine with defined bytecode - first done for Smalltalk. Hotspot optimization technology (dynamically compiling bytecode to machine code) - first done in Smalltalk systems. Full blown object oriented systems with extensive class libraries - first done with Smalltalk (Simula was the first implementation of OO, but the class library was rather small and more targeted at simulation software, Lisps Flavors and Loops were early OO systems, but no class libraries, as they were used to implement AI software). Sorry, but how can you talk about "Smalltalks dream to change the world" and say it was a failure?
Oh, and building things that should last _is_ about building cool things. Only cool things stay around, as people love to play with cool things. Build cool things and you _will_ change the world. Even those boring things that dominate some field usually started as cool - Java _was_ cool when it started it's life as just the JVM with the Java language. It became boring by SUN trying to dominate the software world and adding warts like J2EE instead of coolness.
Don't know how cool it was when Admiral Grace Hopper implemented the first Cobol compiler, though ...