I would make an opposite point: software that's distinctly faster than its peers impresses users not only as faster but also as generally higher-quality in ways one would not think had anything to do with performance.

Users are sensitive to tiny differences in speed, much tinier than most software developers consider significant. For example, there have been many studies showing dramatic drop-offs in web usage with what seem like micro-differences in response times.

A good example is how Chrome is gaining rapidly in market share primarily by being more responsive. There was a fascinating post recently on a political blog I read where the author was talking about switching to Chrome even though he couldn't quite figure out why. I read it as an example of the psychological spill-over effect of performance. (Edit: here it is: http://www.talkingpointsmemo.com/archives/2009/12/the_case_f.... Well worth reading for anyone interested in the psychology of software.)

It's a fascinating conundrum that while hardware has gotten orders of magnitude faster, software has remained sluggish. I know it's fashionable to explain this by saying "but software is doing orders of magnitude more than it used to", but I think this is only a partial explanation; it's also the case that we're leaving huge amounts on the table because of the way we write code. I used to believe in the dream of an ideal high-level language abstracted away from the hardware altogether. That now seems to me a harmful delusion. What we need is a better way to combine high-level constructs with low-level ones. This in turn (veering wildly off on a tangent here...) explains the persistent appeal of C++ despite its manifest perversity. No other mainstream language is even in that game.