It depends on WHICH "real world" you're interested in. I.e., which BOINC project you're going to crunch and even which sub-project.
If you're going to run CPU tasks at PrimeGrid, you'll probably want to stay away from AMD because their implementation of the AVX instructions is very poor. So poor, in fact, that there's no speed benefit in using them. On the other hand, on Intel processors, LLR, PFGW. and the CPU version of Genefer can see as much as a 40% increase in speed when using AVX instructions. ("Your mileage may vary" as this is also affected by memory architecture as well as how many cores are running.) At PrimeGrid, AVX is currently the holy grail, and AMD simply is not currently a player in that arena. So the choice between AMD and Intel is a simple one. AMD's only attraction is that they're cheaper.
Power usage (and heat) also goes up when running the AVX versions of those programs, so I don't know whether the power to speed ratio goes up or down when using AVX. The speed certainly goes way up.
On the other hand, if you're running sieves at PrimeGrid, AVX is not used. However, hyperthreading is useful for sieving (but not for LLR, PFGW, or Genefer).
Most people are interested in raw speed and not all that interested in power consumption, so you may have difficulty finding a comprehensive list that compares the power consumed vs. speed. There's a few data points scattered around on the forums, but no big list specifically talking about TDP vs. speed with actual apps.
My lucky number is 75898524288+1