I've spent some time looking at it... the problem is mostly organizational. getting atoms in rackmount form factors usually adds considerably to the price (and often sticks you with a shitty power supply that means your tiny atom eats 50watts, making it uneconomical) but without the rackmount form factor, sticking a reasonable number of those in a rack quickly becomes a nightmare of a ratsnest.
I'm not saying it's impossible, it just is going to take some work to build a proper rack, and even then, you are probably going to want to take the 'if it fails, cut power and leave it' approach which doesn't do as well with the 'send me your server, I host it' model.
or, maybe they are just really careful stacking it? I don't know. There is a whole lot of demand for cheap co-lo on dedicated boxes.
I'd rather have a VPS on an extremely powerful box with redundant power, etc. It's like being mayor in a rich country instead of being a third world dictator (the dedicated low power box).
The problem is, there are no good standards on VPS provision of CPU time. Memory and disk space are easy to count, but guarantees of CPU are harder to come by and pin down.
actually, the opposite is true. it's really easy for me to give you a certain cpu 'weight' that guarantees a certain minimum and defines who is allowed to burst when it's time to burst; see the free chapter of my book for more info http://nostarch.com/xen.htm - but disk usage, well, disk usage is much harder to limit. I can use IONICE to kindof manage disk bandwidth, but really it's seeks that are the problem, and figuring out who is using more than their share of those is quite difficult.
(Disk space, as opposed to disk performance, sure, that's easy. But space, without performance, is so cheap as to hardly matter at all.)
It's easy for you to give me "weight" or certain percentages, but that's not as easy to decipher as a buyer as the memory, bandwidth, and disk space limits.
Amazon attempt to define things with their "Compute Units" as being equivalent to a single core 1.0GHz 2007 Opteron (or whatever it is), but most VPS providers don't give good demonstrations or examples of the sort of power you're guaranteed to get. Pure megahertz don't help, but even something like "roughly equivalent to a 2 GHz Core 2 Duo" would be a massive help to buyers.
(Even Amazon's compute units can be hard to figure out. How does a 1 GHz Opteron from 2007 compare to, say, a 2GHz Core 2 Duo? These are things buyers are thinking about and looking up endless Tom's Hardware charts for :-))
Hm. well, first, if you want to compare different cpu architectures, that's really hard, because the effectiveness varies massively by what sort of tasks are performed... so this is a task that requires so much fudge factor I'd call it more marketing than engineering, and ignore it.
Hm. But what If I mentioned the minimum and maximum? I can do that... so, for example, on a 256MiB VPS, you are probably on a server with 8 cores and 32GiB ram. (now, personally, I steal 1 core and 1GiB ram for the dom0, but you could squeeze that some, though I wouldn't recommend giving the dom0 less than one core.)
so, uh, you have a total of 124 domains, right? and 7 cores, so each domain gets, what, 5% or so of a core at minimum, and then 100% of the number of vcpus given at maximum.
now, if the provider was simply up front about how much ram they put on a box, the user can figure this all out for themselves (in a xen system /proc/cpuinfo will give you information about the real cpu)
But yes, this could be made much more clear. But even so, it'd probably not result in anything that is usefully comparable to a dedicated server, and really, the performance bottleneck we usually see is disk; if you use 2 or 4 spindles for those 124 accounts, well, disk sucks pretty badly for everyone. But this is the nature of small VPSs.
(so really, giving you specs about my disks might be better; or just making public graphs of CPU and Disk usage on servers. But I have a hell of a time convincing people that anything but the CPU matters, so that might be a waste of time.)
I guess the bigger point I was trying to make is that if you are running on a small-ram VPS, you almost certainly will hit performance problems due to disk I/O before you will hit performance problems due to CPU usage, so being able to compare CPU is of, ah, limited utility.
yeah, if you have some sheet metal skills and a Chinese factory, it's definitely an opportunity. Note, though, the vast majority of people buying in this space are the large corps who don't care as much about money as your brand, so, uh, good luck, I guess?
Honestly, I don't know enough about the costs involved with making chassis to know if there is much margin in the conventional server chassis business.
I'm wondering if there are places looking at hosting little boxes like:
http://linitx.com/viewproduct.php?prodid=12678
Power-wise, they must be on-par with the Mac Mini/AppleTV but seem to be a fair bit cheaper.