That was tested on an Intel NUC with a Kaby Lake CPU. That's the latest generation CPU cores (Coffee Lake is a new die with the same cores), and also Windows 10, so it should theoretically be the smallest performance hit as patching Intel CPUs goes.
In most of the benchmarks, patching the system made performance go down. In several, performance actually went up slightly. That should give some idea of the amount of measurement error. The ones where performance went down a few percent could also be just measurement error.
There were several benchmarks where the performance drop was in the double digit percentages, however, and one where it was as high as 29%. Storage-heavy things seemed to be the hardest hit. So while there is sometimes a considerable performance hit, the performance difference often amounts to a rounding error.
In terms of what CPU you should buy, Meltdown is not in itself a reason to avoid all Intel CPUs like a plague. If you're comparing benchmarks, however, it is a reason to make sure that you compare benchmarks from patched systems to get something representative of the performance you'll actually see. A naive approach of just trying to subtract some fixed percentage as a Meltdown compensation will usually be way off. Unfortunately, this throws out the existing launch day reviews, which is most of the reviews you'll find.