First of all Coreteks 100% knows his stuff and this was the first video i found wanting in some regards.
If we focus on my argument about eliminating bottlenecks, then Coreteks doesn't make a compelling argument(he doesn't for RT as well) based off what we now know. He mentions the custom controller in the PS5 and 12 channels as an advantage over the XSX, but this explains how the PS5 is able to have an impressive raw SSD speed of 5.5GB/s(>double the XSX) and not how it eliminates the IO bottlenecks between what's stored in secondary memory and the actual game code.
...But before continuing, I need to point out something we need to acknowledge that is misleading in Coretek's video. SSDs by default have low latency relative to HDDs, so both systems are going to change how RAM is utilized, the PS5's advantage here is the higher throughput...
In order to eliminate those bottlenecks, one can either rely solely on hardware accelerators or software or a combination of both. The data still has to go through the APU before going to RAM, so software on the CPU can be used to eliminate bottlenecks. Sony used a lot of custom hardware accelerators in the APU i.e they rely much on hardware, in order to spare the CPU. On the other hand, MSFT used a combination of Software and Hardware accelerators. The issue is, we do not know enough about MSFT's IO system, but what we do know so far is impressive and on par with the PS5's in terms of eliminating bottlenecks. MSFT is claiming that they only have File IO overhead equivalent to one tenth of a zen 2 core.
And my issue with Coreteks' video is that although he claims to be privy to some extra information, he's unable to clearly articulate(without revealing) what bottleneck the XSX may have. Instead he only mentions the public information from Cerny and infers that the XSX won't be able to eliminate these bottlenecks(which we now know is not true). The only thing he mentions about the XSX that was not publicly available at the time is that it has WDP compression but he doesn't mention BCPACK(which is the most important one) which at the time of his video was not publicly known to be present on the XSX.
Of the five bottlenecks presented by Cerny, the most important ones are File IO(including mapping) and Decompression. From what we know as of today, both systems eliminate these. For the others we'll find out but for example DirectStorage should be able to address DMA since it will be doing it for virtual RAM. So we still have a lot to learn.