...
OpenSSL private keys aka. 256-bit random numbers by themselves will not help with reducing the number of keys, however it is estimating the bias (as in - empirically, some bits will appear in a certain state more often than others - when several thousand random numbers are generated) in OpenSSL random numbers this estimation method draws its power from.
Also this doesn't work if your sample only one private key, that's why I mentioned that thousands of random numbers need to be sampled. My calculations get away with just 1000, though.
...
I recently did a bit allocation study on generating private keys via /dev/urandom. More than 100 million keys were used for the test. The bit allocation was about the same = 0.5
Maybe the old versions of OpenSSL, which were probably used in the puzzle creation, had some vulnerabilities and gave some deviation?