I think it's a weird thing that UAD Precision Limiter doesn't do oversampling. However it does do, two different look-ahead modes which are selected using A/B switching. You can really push Sonnox Limiter hard and it's doesn't really distort the sound. The whole point of the premix stage before it reaches the enhancer section is it uses internal compression before sending it into that stage of the process. It actually does quite a lot of things. 
 
I'm not 100% sure how they came at these figures and what null test they used exactly. Not being an audio engineer I will take their word for it.

Another thing is why do people insist on using s**** lossy formats as they test ground. Sure I listen to Multimedia formats, 256K+ and FLAC typically.
 
 

.

.

i think my tc electronic meter might have a different measuring system as every limiter i have- sonnox, ozone, psp are measured at 0.1 dbfs "true peak max" . the new sonnox limiter is pretty cool to watch because it details the isps as it works and gives their level. the meters on the limiters themselves all show 0.0
 
tc electronic have their own limiter but it's a bit different. you drag your file onto the app and it looks over the file and adjusts accordingly. would b interesting to own that to see if it conforms to the tc electronic standards. 
 
makes sense all the limiters show different readings on the same material when they were perhaps calibrated to different meters during coding. can probably guess that the voxengo guy also tests with ozone metering as elephant came out to -1 exactly

You have to be hitting around 7dB RMS before you are heading over true peak values. I mean some recordings I have are +3.2dB true peak whilst the LU level sits around 8 8 or 9. Anything after 0dB will be basically amplified or distorted unless coloured by the limiter as far I am aware.

mister happy 
After, all the only way to constrain an inter sample peak is to modify the relationship between the samples before and after it. How do you arbitrarily modify sample values with varied rates of scaling and maintain a concept of accuracy? It seems like a compromise is inherent to the process.
  

Which is why I pointed out above that "more accurate" likely means "more distortion" in practice - basically you are just setting the threshold a little lower and perhaps even deliberately distorting a whole series of samples just to address an isolated peak that wouldn't be audible on it's own.

.