Considering the list seems to contain both "big" and "large", my guess is that there's quite a bit of overlap in the words used because they expect the 3000 are generally known and can be relied on. This means that if we were going to optimize for size, we could probably get to a much small number of words, and use those to define the others.
I didn't go searching, big was literally the first word on the list I read after going down a few pages, and I wondered about large, so I searched for it. I just looked a bit more, and there's "child", "childhood", and "grandchild", which while not the same problem, does illustrate that they are fairly liberal with their inclusions because they appear to want to use the minimum vocabulary to define something idiomatically, which is a slightly different question than what's the minimum required.
This problem actually seems to share a lot in common with database normalization.[1]