@rml
If you were only interested in computers as a brute-force calculating tool, or interested only in the business side of software, you aren't interested in Lisp because it lost out to languages like Python, JavaScript, C/C++. So I think any Lisp will only attract people who are interested in lambda calculus and/or programming language theory, and/or maybe people interested in dependent typing, like anyone who has run across the work of Dan P. Friedman.
I assume it is not just me that the reason Scheme is appealing is because it is a well-designed minimal Lisp, and just being able to understand, from a pure computer science perspective, the implications of what a "well-designed, minimal Lisp" even means has already narrowed down the pool of potential converts to a tiny minority.
But the fragmentation is still the biggest problem. The absolute first question I had when I wanted to get started with Scheme was, "which implementation should I use?" And immediately it becomes clear that once you have picked one, it isn't easy to just switch your code over to some other implementation if you feel like the one you picked first is wrong. So there is soooo much pressure to pick the right implementation on your first try. That alone I think scares too many people away. I didn't run away because I was already committed to the idea mastering a "well-designed, minimal Lisp."