Agreed.

All too often I find that when people say that they 'denormalised for performance' they actually mean 'didn't feel like normalising completely in the first place'

Denormalising for performance means that you've normalised fully, have then done tests on large volumes of data and have found that the fully normalised design is not capable of meeting the performance requirements. It does not mean "Oh, I think there are a few too many tables, let me denormalise a bit"

Oh, and I don't mean normalise to Domain-Key normal form. 3rd or BCNF is usually enough, maybe 4th if you've got odd tables that show that particular data anomaly.