For node, you simply want the highest overall performance. The Xeon line-up is structured that there is never a model which has same performance at lower or higher-core count, it goes like sinusoid instead. Currently the highest-performance yielding models are the 20-22 (there is even OEM 24 version) cored ones.
For workstation though, it's worth sacrificing a little bit performance and go for the lesser cored, higher clocked versions for few reasons:
1) Avoid processor groups like plague. (Non-server) Software developers didn't get the memo Microsoft introduced such feature. Corona and Vray work fine though.
2) AVX under-clocking base-clock and turbo bins. AVX instructions further lower the base clock, which Xeons already have rather low, and even the turbo clock is lower in such case, doesn't make big difference for rendering but it could in some other cases.
3)The weird issue with metro start-up in Windows 10. I honestly have no idea what to do with this, but I didn't had much time to investigate yet. It's ok for node, but very problematic for workstation>
4)You want the highest turbo bins. Currently most of the line-up has pretty flexible and powerful turbo bins, so even the low-clocked Xeons can go from 2.6 into 3.5Ghz for single-threaded processes. These makes v4 Xeons excellent workstation CPUs. Certain ES/OEM versions don't have this ;- ) So unless you're buying "authentic" or latest QS release of CPU, don't use them as workstation.
Because of this, it almost doesn't matter what kind of Xeon clock you get, because the turbo is so flexible (this wasn't the case in previous generations). Higher clocked Xeons in v4 don't make better workstation, pretty much all the line-up works the same, although some go up to 3.8+ Ghz in single-core turbo bin. It's because of the above issues you want to avoid it for WS.