You are right! :)

But an hash (DB_File) with 2/3 millions words+phonetic codes is around 150+Mb.
So to gain some speed-up you have to twice the database: one that has keys as phonetic codes and the second that has keys as correct words.
And this is not always a good thing.

But in this particulary case, with only 100.000+ words, your suggestion is ok :)