Firstly, don't blindly read any guide (whether provided by Magento or not) and try to ascertain server configuration and server specification for your own deployment.
Their tests and results are entirely specific to their test conditions and not applicable to your own store.
Every Magento store is different
Beyond that, selecting hardware for your deployment isn't something anyone here can do with the information you have provided.
You haven't provided nearly enough information for a useful response.
- How many visitors do you have per day peak
- How many visitors do you have per hour peak
- Do you offer digital downloads
- What amount of transit are you currently peaking at (in Mbps)
- What proportion of web traffic is SSL and none SSL
- How long are you prepared to keep stale items in a cache
For a reverse proxy, you need ...
- Proportional amount of RAM for your cache stores (min 4GB)
- Standard disks (disk I/O is not relevant)
- High end network card (to reduce interrupts, improve latency & throughput)
- Proportional amount of CPU cores to traffic level
- Optional. SSL hardware decryption
Don't bother. At the moment, your deployment already has flaws
- 2 single points of failure
- Bottlenecked MySQL throughput over network
- Too few cores in web server
- Too many cores in DB server
- Over specified HDDs in web server
- Under specified HDDs in DB server
Believe it or not, Magento runs better on a single machine up until you hit the limits of vertical scaling, because the bottleneck of MySQL being on a high latency remote machine constrained over whatever you network speed is - is far higher than having it on the same system and having RAM speed throughout.
Don't get me wrong, there's certainly a tipping point where multi-purposing 1 machine creates its own contention issues; but 32 cores is a fairly small ceiling of load.
If you still press ahead
A Varnish cache for a low traffic site (anything under 200k unique visitors per day) doesn't usually need more than 2GB RAM and a 1 core CPU - with low end HDDs. But for you, it may as well just be on your web server anyway - it wouldn't draw any resources, and by putting it on a third machine - you'll now have 3 single points of failure and even more network activity bottlenecking your throughout.
Take that paper with a pinch of salt - it is a marketing exercise from Peer 1 and the majority of it is not correct.