There's no benefit in trying step 2 because the gibberish input is as likely to be a close match as a sentence that makes sense.

A sentence that starts 'Be' (001000010...) and the same sentence changed to 'He' (001001000...) would produce very different hash results by the Avalanche Effect ( http://en.wikipedia.org/wiki/Avalanche_effect ). So having similar letters in a similar order or in similar positions, or using similar length words, wont help.

Isn't that the whole point of this competition?

The point was to get a hash with a similar quantity of 0 and 1 bits in the output as their hash, not to guess their sentence (they said what it was).