You base your entire argument around an idea that the web hosting sphere is somehow fixed, solid and unchangeable (or so it reads). So, a host offers 1000 clients space on a server assuming that they'll all happily co-exist and use far less resources than advertised. Most of them will. Some of the clients will strike it lucky and have popular websites that use up more resources than their "fair share" but less than they were advertised.

Now the host can take two paths; boot off the resource hogger (with a made up excuse about breaking the AUP) and still have 999 happy customers, or make allotment for the customer and give them what they paid for (or at least what they need) by perhaps migrating them to another server.

Some hosts will take the easy, first option. These are the oversellers that you want to avoid. Some hosts will take the also fairly easy second option, and these are the oversellers that there's no reason to avoid. It strikes me that overselling in and of itself isn't a bad thing. It's what the individual hosting provider does with regards to people using more than the "average" resources that really matters. Hold on a minute, lets back track. Overselling in and of itself isn't a bad thing.

From what I can see (which isn't a lot, stupid myopia) the entire argument boils down to don't go with a bad host, in it's simplest form. I fail to see your intended point that all overselling is bad.

Interestingly, but completely off topic, I get only 75,000 results for "dreamhost sucks" in Google [cf, Jules' 137,000]. I also get 894,000 results for "google sucks". D'oh guess it's time to stop using the latter's services.