"Otherwise I don't understand the point of the whole thread and am quizte happy I didn't try to waste time on coding an even faster solution."
That seems a bit harsh. Everything here is still usable from Inline::C when pure speed is wanted. Since I made a "clone" of an existing module, it seems having it work like that one, but still quite a bit faster, is worthwhile. Apparently, it's not just Standard & Poor's using letters and values > 10 for this kind of stuff. Looks like OpenMRI and The Regenstrief Institute do as well.
I can either make another module that's purely about speed, or see if your suggestion above pans out for this module.