I've seen a CMS brought almost to its knees because the previous owner of that IP had a site that had lots of distinct pages on it. Since every page in the CMS was stored in a DB it took a DB lookup to find out whether the incoming URL existed or not. Caching/varnish wouldn't help as there were hundreds of thousands of different incoming URLs and none will be in the cache because they don't exist.
About 20% of the hits to one site I look after are 404 because they're from the previous site hosted on that IP address. Luckily the vast majority of URLs have a specific prefix so it's a simple rule in the apache config to 404 them without having to got to disk to check for the existence of any files. It still counts against my bandwidth utilisation too (both incoming request and outgoing 404).
Presumably to avoid being plagued (in terms of load and bandwidth costs) by the numerous crawling bots looking to update their caches of pages that no longer exist on those domains.