>I'd go with VMs.
>
>(a) If you have a dedicated server and it goes down your out of business. Most cloud services will automatically spool up a replacement instance.
>(b) Most cloud services can be configured for a guaranteed number of cores (Amazon has 'Burstable' options where credits for CPU under use can be used against high usage periods). AFAICS bottlenecks are more likely to be at the NIC level - but that would apply to dedicated servers as well.
>(c) If you use VMs it's easier to scale up/down as demand dictates and you only pay for what you use.
>
>My 2c.....

I'm with Viv, I think that VM is better here. Normally the hardware used for hosting VMs have many CPUs, and a VM can be assigned 1, 2 or more CPUs in exclusive, so yes, it is possible to have "exclusive use" of CPUs if, in example, the ISP have 20 CPUs in a Server and assign 2 or 4 of them to you. If you need more CPU, then you can pay for more.

Anyway, 2 years is a good time interval to try this and get conclusions, without risking too much.

I think that the question you can make is this: How many VM hardware can you get, in the worst case, using the same money that you could use with dedicated hardware?

Regards!