First of all, the thing you are replying to isn't in reply to OP's article. It's in reply to a particularly stupid comment.
Secondly, the HN crowd doesn't seem to consider that when selling to large corporations the Dell, Lenovo, etc. "overly complex" approach makes sense -- that there might be a reason that apple isn't the #1 computer manufacturer, especially for institutional purchases.
Fourthly, before talking about "mostly waiting on user input" you should consider the research. ("1.0 second is about the limit for the user's flow of thought to stay uninterrupted") start with http://www.useit.com/papers/responsetime.html
Fifthly, you seem unfamiliar with how scientific computing works. Here's the process:
1. Hypothesis (usually but not always done by professor or adviser)
2. Find a dataset, or multiple datasets that should be able to falsify #1
3. Code up an initial test in R or Mathematica (or Matlab if you don't like FP) -- does the correlation even exist? How well do the off-the-shelf methods work?
4. If you think you can do better, write it up in C++/Java, testing as you go
5. If the process requires a lot of CPU time for near perfection (NN / RBM / GA / GP), run task on server for use in your publication (graphs / discrimination percentages)
Only #5 is really effective to put on the server... Working with Mathematica remotely is a PITA with non-university hardware (licences), and remote interfaces have annoyingly high latency IMHO.
Sixthly, I do like fiddling with knobs, but to use your own turn of phrase back at you... just admit it: you didn't think before writing what you wrote. You're not analyzing and trying to understand anything in your post, only stating an opinion backed up with no facts. Like a child wandering in to a university lecture and announcing that there are no negative numbers because your teacher told you so.
That you were not downvoted to oblivion before I saw this is merely symptomatic of HNs irrational love of apple products.