Hello
There are double strlen() invocations in getstrvalue(). It's not about
negative indexes – although they have been optimized too – but about
dupstring() and v->end, v->start verification code at the end. Actually
a single strlen() call is required. Attached test script runs 1765 ms
for optimized Zsh, 1980 ms for no optimizations (minimum obtainable
times):

strtest() {
    a=""
    i=$(( 4000 ))
    while (( i -- )); do
        a+="a${a[1,200]}"
    done
}

I've used signed size variable but the code did that already. More
optimization is possible – dupstring() could do strncpy(), copy only
requested number of bytes, e.g. first 200 out of 10000, but the trailing
\0 nuances made me wait with this change.

-- 
  Sebastian Gniazdowski
  psprint@fastmail.com