It drives me crazy.
My "favorite" anecdote when it comes to the inefficiency creep:
One of my favorite text-editors is an Amiga editor called FrexxEd. I ported it to run on AROS, a portable re-implementation of the AmigaOS API. AROS can run hosted on Linux with X.
On my laptop, booting the Linux hosted version of AROS, that initializes the full GUI etc., with the startup-sequence (a simple shell script that gets started on boot) set to start FrexxEd at the end lets me "boot to FrexxEd" faster than I can start even the terminal-only build of Emacs on the same machine...
http://www.filewatcher.com/m/pcc12c.zip.175811-0.html
It came with a demo game, which was a Caterpillar clone which ran in text mode. (Unfortunately the archive above doesn't have the source, only the executable.) This game used loops for timing.
When I ran it on my high-end 486, at a whole 66MHz, the screen would flicker madly for a couple of seconds and then the game would exit. In that time it had run through all three lives with no user input.
That 486 was at least an order of magnitude faster than the 4.77MHz 8088 that the game had been written for. And modern machines are a good couple of orders of magnitude faster still. I wonder how long that game would take to run on the desktop I'm typing this on...
It's also funny and a bit sad that despite these extreme "simplifications" of UI, software still seems neither faster nor smaller...