Bill,
The ShopBot machines could be ruled out immediately. Their construction is far too flimsy to ever allow any serious machining, even on aluminum. While it would be possible to eventually end up with a mold via nibbling and gnawing with a shopbot or similar machine, the lack of rigidity would leave you with surface finishes requiring more hours of polishing and handwork than any sane human would ever want to spend on such a fun task
What you're wanting to do can be done by any number of vertical machining centers. In looking for a used machine, past the obvious mechanical health and operating condition, there are two major criteria I'd look for.
First, get a fully enclosed machine. Its essentially impossible to keep the chips or coolant contained within anything less, particularly when cutting aluminum, due to the high spindle speeds and high coolant volumes necessary to keep the cut clear of chips so that the best surface finish can be maintained. I've cut some similar parts on an open cnc mill, and for the entire duration you have a fulltime job with a mop and a broom in trying to keep the mess contained. It gets old, real fast.
Second, you need a control with adequate block processing speed to be able to run a typical mold cam program without starving for data. Mold programs are typically made up of thousands, or even hundreds of thousands, of short linear moves, and a lot of older controls are incapable of processing the lines of code fast enough to maintain constant feedrates. Every place where the axis motion stops due to data starvation will leave a place in your mold that can easily be seen and felt, leading to lots of hand finishing work that wouldn't be required with an adequately fast control. You can't get around this situation by slowing the feed to allow the control to keep up, because all that will happen is that the general finish will go to hell, along with your tool life, due to inadequate feedrates.
There would be several dozen different brands and models of machining centers in the used market that would meet the above requirements, so you have plenty of options on what machine would do the job.
Its my opinion that you should think about getting more quotes for your work before making a final decision to buy a mill. Going under the assumption that this is a relatively low tolerance mold where surface finish is the main objective, and that it doesn't have a lot of features you haven't mentioned, such as a bunch of gun-drilled coolant passages, etc., I'd agree the $20K price sounds ridiculous. However, a big drawback to doing it yourself is going to be the question of how rusty your knowledge becomes during the months when your mill sits idle between jobs. I work by myself most of the time and have a lot of different machines. When I start to run a job on one I haven't used in several months, it requires a re-learning time before I'm back up to speed. This is true even with manual machines, but is about ten times more evident with cnc machines. The same would be true with whatever CAM software you choose, and in most instances that re-learning after a long layoff can take lots more time than you'd spend getting back up to speed on the machine itself. Assuming you can furnish a shop that's quoting the job a solid model from Rhino, and assuming the shops quoting the job have CAM software capable of importing and directly processing a solid for generation of the code, I'd have to agree with Ted (smallshop) that something in this picture is way out of line. Shopping for a supplier might pay lots more dividends with lots less headaches than any other option. If a person allowed $2000 for the material for your 19x27 mold, which would be high, that still leaves 4.5 work weeks at $100/hr to make the mold. Seems more than a little steep to me.