How bizarre. (Ooh, baby)
I'll, uh, show myself out then, yeah? Yeah. *cough*
Number crunching with MATLAB on big datasets can eat a lot of CPU time - it's better not to wait longer than you have to, and throttling will hold you back. Ditto video transcoding and other kinds of scientific computing (like oil reservoir simulation...), or distributed computing projects like BOINC. But for most use cases you're right: there's just not that much stuff typical users do that needs to grind on a CPU for even minutes at a time, let alone hours or days.
My sympathies on the RAM. That's the most evil kind of corner-cutting to save a couple of bucks.
"A car is just a big purse on wheels." -- Johanna Reynolds