I'm the total opposite of a server guy, but I think your tests may be in some way flawed. At work we have a site that got high traffic for a few months on a dedicated box @ softlayer -- and it keeled over, unable to handle the outgoing data.

Moving the site to a EC2 large completely removed the problem.

I realize this is completely anecdotal without much detail (well I know the softlayer box runs about $200 a month, so it can't be their worst thing) but it does make me wonder.