OK, so I'm looking for a server that will run a VST host (fst or dssi probably) and perform a vocoding style effect using a plugin on samples of sound ranging as long as about 5 minutes long as the longest. It will take unvocoded samples, and return them with the effect applied. This would probably have a maximum of 2,000 users (total with access, not 2,000 people using it at once), who would be using it casually.

Is any Linode plan capable of this? If so, which plan would work about right? If not, does anybody know what server specs I'm looking for?

I'm hoping a Linode could do this since I personally am already a happy Linode customer, but my business might need something more powerful.

I am not sure whether VPS is that good for CPU intensive taks. Surely people co-located with you may be unhappy at times

Have you considered EC2 (maybe High-CPU version) - booted on demand to perform the calculations?

Yes, like Mekk suggested, Amazon's AWS application stack is exactly what you need for this type of batch processing.

The process would look something like this:

1. Audio samples get uploaded to a web app running on a "front end" EC2 instance.
2. This server transfers the sample to storage in S3, and then submits an item in an SQS queue.
3. Depending on the number of pending jobs in the processing queue, one or more processing nodes in EC2 spin up, grab items out of the queue, process them, then upload the result to S3, adding the processed item's metadata to some sort of DB.
4. The front-end EC2 instance serves up the end result via web, email, etc.