Hasn't this video been posted before? Anyway: In general I can probably pull out hundreds of samples per day that each given solution doesn't detect. That may sound horrible, but given that those are from 400k a day puts things into perspective.
We do miss samples occasionally, yes. Every single solution does. Some of that is in public tests, others are on actual user systems (which is the worst case scenario for us and is thankfully exceedingly rare), and others we notice during our normal in-house sample testing. Trying to find any kind of pattern ("this product is getting bad!") based on limited insights into some of those misses is kind of silly. I would even take official tests with some grain of salt. I mean, we literally throw tens of thousands of dollars at testing labs each year and even I can't tell you what exactly is contained in those test sets. We will only ever see the samples we missed. But there is almost no transparency into where these samples are from, how they were obtained, whether those are actually different malware families or just hundreds of the same family, or how any particular sample ended up there.
One particular case to give you an idea was one particular VB test, where we ended up missing like over a thousand samples. That looked obviously pretty bad. Turned out, all those 1000 samples was the same virus that had infected over a thousand different files, which all somehow made it into the test set. So a product that missed two samples, may have been worse than us, even though we missed over a thousand.
So while I appreciate people's efforts to make these videos (I actually enjoy watching them a lot, it's a guilty pleasure of mine), it's just one isolated data point in a vast sea of millions of data points collected all over the place.
In my experience, and feel free to dismiss this if you wish, the best long-term indicator is looking into tech communities. Places like here or BleepingComputer or GeeksToGo or Trojaner-Board. Places where victims of malware show up. Then see what kind of products they use. You will see users of certain products showing up there quite regularly, while others don't, even though both products have comparable market shares. That's like the closest you can get to real life performance insights.