A search that had to first convert every single tiddler to to non-accented words would really bog down the search, IMO.
I'm thinking a better way would be to modify the saving mechanism of the edit tiddler so every time a tiddler was saved it would go through the contents, change all accented words to unaccented, compile a list of unique words that have been made unaccented, and append them either at the bottom inside comment tags or inside a special "accented" field. This would only add a fraction of a second to the saving process. Obviously this would only happen in systems where people had opted to create unaccented search text.
The search process would then know to check the "accented" field when doing a search for deaccented (?) text. The actual size of the TW would only be bloated by number of words that had been deaccented, which presumably be a small subset of the total, so the search would still be very fast.
Mark