Having some experience commiting to back-end and middle-end of certain popular programming language compilers - it's not always so simple. The length of the code doesn't necessarily map to the length of machine code. It's a good rule of thumb, but it might not apply in this case.

These loops are idiomatic and easy to identify by SSA optimization rules.

There's a comment somewhere here that says it's the fastest solution. I haven't really built it with that in mind, but if it is, I bet it's because it's easy for the runtime/compiler to apply optimizations and vectorize the loops.

Just looking at them they seem ripe for some optimized SIMD bit magic implementation.

I don't think it's over documented, although certainly it could be shorter.
That can also imply less perf

point is the length test shouldn't be on the passed string, but on the final hash-tagged string.

It can be way longer than 139 characters. Remember that the spaces are not going on the hashtag. Something like this "a" + " ".repeat(200) + "b" should be valid.