Sure, there are some worse/more limited predecessors – my design was partially motivated by a desire to improve upon these.

For instance, that ML-derivative you are using for your examples

• very likely still has if then else in the language, thus making it not unified
• desperately tries to emulate functionality with guards that simply comes out of the box with my approach
• relies on the ultimate hack of “match on unit”, because match is very limited in which coding patterns it can express

Also, none of the examples are “more clear” or “have less magic”:
Maybe they are more “familiar” to you personally, but that’s about it.

Too me they just look clunky, full of accidental complexity and trying to work around a poor/limited language design.

That –at best– gives you the same performance.

EDIT: Ok, I misunderstood – you meant the performance of “case insensitive in kernel” vs. “case insensitive in userspace”. I get your point now.

How would that happen?