I found this article very useful, but I think that it goes a little too far. I have seen a few comments which I agree with regarding the default use of clustered indexes which of course can be a source of performance problems when used inappropriately, but I thoroughly agree with the gist of the article.

I would have thought (other than tables with a single lookup row, typically used for global system values) any table that cannot be keyed in some form is more a sign of bad design. How can you single out a particular row of a table (say for deletion?) if you cannot uniquely identify it?