tnt, you made some good points there, but looking through each one, I think I covered them already in my last answer to 9600.
The quick version is that sure, there will be lots of high-speed embedded applications of Parallella (your project provides a great example), but there won't be many as a percentage of Parallella users I believe, because most people will be using the board for parallel computing that exploits the Epiphany array alone. (We'll have to see what the future holds.)
Between that great majority that attaches no daughter cards at all and the ambitious bespoke projects like yours that will mostly have populations of one, there is a middle ground for those who want to attach a commercial or OSHW resource module in a plug'n'play manner with the relative user-friendliness of adding plug-in cards to a PC, but with the far superior engineering advantages of basing expansion on Ethernet.
It's a more distributed concept of how computers can be structured, more in tune with modern-day networking and with the needs of clusters to access distributed resources uniformly, and it has all the cost and engineering advantages of riding on the back of existing technology.
It's also worth pointing out that industry has already recognized these huge advantages, and that is why industrial fieldbuses are very commonly based around Ethernet as the underlying fabric linking distributed resources.