Unfortunately, this result might just mean that there's a correlation between noise tolerance and performance. It would have been better if they had studied objectively measurable noise levels instead of subjective acceptability of noise levels.
I certainly feel there is a correlation between _tolerance_ and performance.
I find hearing a muffled conversation is actually worse than hearing a plain conversation. My pet theory is that either way, the brain has to commit some resources to parse the voice into actual concepts, but if the conversation is muffled it has to do even more work to 'clean the signal'. Using resources that are taken from your concentration ability.
You can consciously try to ignore the voice as much as you want, your subconscious still have to parse it. At least i've never been able to avoid it. I think that thats why white noise or music does not affect concentration nearly as much as a conversation of the same Db-level.
You can try to train yourself not to parse conversation, to think of it as a slightly irritating set of noises. I think we all try to do this to some degree – it may be one reason why programmers are famously uncommunicative when we're in the zone. I mumble and wave people away when I'm trying to sustain focus, and even if I deliberately break focus I tend to take a couple of minutes to snap out of it and get back into the flow of conversation, and in the meantime I struggle to produce words.
But, like you, I'm not convinced that consciously ignoring words really works. Your brain is still parsing them at some level. It's like the studies which show that noise, in general, is stressful, even if that noise is just white noise. Your brain is trying to find signal in the noise. It never really stops trying. The more noise there is, or the more structure your brain detects in the noise, the harder it will try.
Studies have shown, and my own experience bears out, that voices are hardest to tune out, and that structured music is also relatively distracting: Your brain parses music too. Ambient music, which is deliberately designed not to attract conscious attention, works better for me. White noise works, too, to some degree, but silence is still best.
Of course, some people are conditioned to have noise all the time. These people are both lucky and unlucky, in my opinion.
Socially, and evolutionarily, we've adapted to pay attention to those who matter to us. That means boss, co-workers, and especially family and community partners. We're also adapted to picking out signal (especially speech and movement), which is why muffled talking (more effort to hear and understand it), and any moving video, can be so distracting.
It's why working in a "busy" spot (cafe, other public area) in which those around you are not part of your immediate society can be easier than working in an otherwise quiet location with someone close to you (and why getting to know people at the cafe "ruins" it as a working location).
Know your inner animal.
I think it's less tolerance and more focus. Workers that can focus appropriately can store the requisite information to solve the problem in their mind. Those without focus will introduce defects as information not task related is encountered and processed.
(1) Some workers who had previously produced defect-free code started producing code with defects as the noise level increased.
(2) Some workers who had always produced code with defects started complaining about the noise as well.
In one of these cases, keeping noise levels down would make a difference. In the other it would not, and the numbers as given don't tell us which scenario applies. They also don't tell us the relative numbers of noise-tolerant vs. noise-intolerant programmers. If those affected by noise are a small minority, then an employer might make the choice to maximize office density and let those affected "select themselves out" by going elsewhere. OTOH, if just about all programmers are affected, that's not much of an option. If the authors reported that the number of programmers producing defect-free code went from X% at 20dB to Y% at 50dB to Z% at 80dB, then that would be much more useful. Similarly, these studies don't seem to measure time to task completion or ability to complete multiple tasks per day, which would be other factors in such decisions.
BTW, this is more than a theoretical issue for me. I'm working at home today because the noise level in my office - due to high density and low partitions - has become utterly intolerable. Just about everyone wears headphones all day, presumably because they feel likewise. I'm a bit disappointed that one of the most-cited studies in this area does not really support any useful conclusions that can be brought to management.