How can you trivially conclude that its would be easy to provide the same speed as Blomberg terminal ?
If I am not mistaken even shaving off fractions in terms of time has a lot of value.
And maybe you need 25 million lines of FORTRAN code to achieve that efficiency ?
This sentence makes me dismiss the entire article:
"The trading model and real time issues are just technical problems. Google is smart enough to solve them."
And I have written mission critical Fortran software for big telcos' and tust me nothing contreates the mind when one on Vint's reports nudges you and says this had better be right or we are both looking for anew job.