I took a brief look at the code, but I fail to see how this transformation could introduce any additional secrecy. If the randomness used to define the polynomial is good, then Shamir's secret sharing provides information theoretic secrecy (no matter how the secret actually looks like). What Ricky points out in his answer seems reasonable, i.e., to provide some means against the malleability of the secret sharing. Since Shamir's approach is linear, one can "update" the shares to "update" the secret and if you add an additional layer of diffusion to the secret before sharing such "updates" are no longer possible in a straightforward manner. Note, however, that this does not influence the secrecy of the secret but against active attacks on modifying the shares.
Another issue which I initially thought could be the case comes below.
If your secret is larger than what
can be represented as an element of the used field, then computational secret sharing (CSS) can be used, but it is not implemented in ssss (it's only linked as an alternative at the bottom of the page).
Basically, with CSS it is possible to make shares in a secret sharing scheme shorter at the cost of losing the perfect
secrecy guarantees provided by Shamirs approach. One only achieves computational instead of information
theoretic security.
The idea is to use any IND-CPA secure symmetric encryption scheme in the following (obvious) way:
Choose a random secret key $k$ of a suitable symmetric encryption scheme and encrypt the secret $s$ using $k$.
Then one uses polynomial secret sharing to compute $n$ shares $s_1 ,\ldots , s_n$ of the key $k$.
Note that the field $\mathbb{F}$ used for secret sharing here can be much smaller as in Shamir’s original approach,
as you only need to represent the key $k$ as field element and not a potentially large secret $s$.
Note that if you use Shamir's secret sharing directly on $s$, then $s$ needs to be represented as a field element (as the constant coefficient of the polynomial) and also the shares will then be of that size. If your secret is large, then CSS is an efficient alternative.