From the comments section of the article:
The XMOS desgin suffers exactly from the same problem as the Inmos Transputer: Although the basic idea is very good (CPU having inter-process communications facility built-in), the implementation is very British: Instead of integrating a differential driver into their chip and using the latest process technology, they have a substandard-performance communications links yielding about 100Mbit/s.
With differential signalling, they could certainly achieve 1Gbit/s or more. There is a technology around called 10G Ethernet.
The Transputer had exactly the same problem...
Which pretty much nails it for me
That the chip itself cannot access system ram (or afaik anything else), runs essentially on its own and the open sourced dev tools only allow you to operate part of the chip... I just can't understand the necessity of it on a desktop system beyond being a kinda cute talking point. It's a solution looking for a problem.
The article is so much crap it isn't really worth talking about.
