It seems to be something more funny than useful. These "words" don't add anything to simply using Base58 (and perhaps adding some huffman coding upon it to make it auto-error correcting).
They add pronouncability and case-insensitivity. Base58 is designed for machine consumption, RFC1751 is designed for human consumption.
Two examples:
FEED and FEET. Their "distance" is one character (D and T) AND they have a similar sound (D and T are both dental consonants).
DIME and DINE. The only difference is a single segment. Are you sure an OCR will distinguish between them?
Another problem: the length of each "symbol" is variable (A, AD, ADA are three legal symbols), so the space is an important separator. ADADAD could be AD AD AD or ADA DAD.
This isn't intended for OCR. There are better solutions for OCR. This is intended strictly for human use. I don't see how a human could lose a space. Humans do not think "therapist" is in any way similar to "the rapist".
In any event, the coding scheme doesn't have to be RFC1751. That's just nice, IMO, because it's a standard. But if people would prefer a BitCoin-specific method, maybe one that was more language neutral or added error correction and not just error detection, that's perfectly fine with me.