There is actually a good reason: it's precisely because they bring in no semantic baggage. Although conses are usually used to build list structure, they don't have to be used that way; you can make binary trees or even arbitrary graphs with cycles out of them (creating directed cycles requires side effects, of course). The accessors 'first' and 'rest' are appropriate only for conses used as list structure. Even 'left' and 'right', which are less semantically loaded than 'first' and 'rest', wouldn't be appropriate in all cases.
So, what would be a better name than 'cons'? 'List' is not right; the empty list is not a cons. Scheme uses 'pair', which is admittedly not too bad, but I think it would be more appropriate for an immutable pair than a mutable two-element struct. (I actually think there should be two types, immutable pairs and mutable conses. In such a Lisp, conses would probably be fairly rarely used, and much more of a historical relic.)
And to echo what outworlder says, it really just isn't that big a deal. Yes, Common Lisp has some historical baggage, and a lot of things we would change if were designing our own Lisp from scratch — I assure you every Lisper has their own list. But CL rationalization projects, of which there have been several over the years (and, no doubt, more that I'm not aware of), don't catch on. The reason is clear: the CL community is already "small yet remarkably fragmented" as I recall someone putting it (Eric Raymond?); a rationalized CL would just fork off another sub-community, if it caught on at all.