It depends largely on the specific customer environment and development, that said, I'd recommend the fastest, most power consuming (yes, this is not green) CPU, Intel CPUs, Xeon E5, likely the fastest RAM available, and if the document uses 6 Gb of RAM, plus the general rule of the 10% per user, up to a max of 30 concurrent users.
Depending on how complex the model is, how complex the document is (lots of simple objects or a few very large and complex expressions) and how the reload and distribution process has to be, the specific answer will vary widely.
I'd recommend you as well to engage with your nearest QlikTech expert or partner to get a more accurate answer.
One important issue is the price... I guess if we would take 1 TB RAM and 4 quad CPU, it could reach 40,000-50,000 $, which is a bit expensive. So I consider to be content with 250GB RAM with the ability to be upgraded up to 1TG, and maybe 2 quad CPU. I still need to figure out the storage volume of the Hard Disk. As well I consider purchasing a Flash disk for development purpose, because I noticed that for large QVWs it takes a lot of time of being loaded to memory and being saved.
Our Qlikview models are tend to be complex, there are a lot of SET ANALYSIS expressions... So I guess I'm aiming to the minimum specification.
Disk usage is important to store the QVWs and QVD files, so it all depends on how much do you expect the application will grow each day. We are deploying large documents in Windows NASes or SATA HDs, and you are totally right in regards to disk read/write delays.
How often do you need to reload the document? The more often, the larger the disk will have to be. Same with the number of QVD files you are reading from and storing into.
As with the processor and memory: the faster the disk is, the lesser the document will time to load into memory, save, reload and distribute!
As a side note, QlikView is not always about having a lot of cores (actually there are cases depending on the hardware where less processors perform far better than more processors!), or RAM Gb, but having fast cores, fast QPI and fast memory access, plus some other configurations you can check in this doc and make sure your hardware complies.
And of course, it all depends on the user experience. Some users find OK when a chart takes 20 seconds to render, whilst others find unacceptable when a chart takes longer than 10 secs...
Again, engaging QlikView experts in your partner or QlikTech may be a good idea, as they are closer to your customer infrastructure and scenario than any of us can be.