When r2 is more useful than r in linear regression?

Multiple R is the correlation between y and y^ in a regression model. It is always non-negative, but has no nice interpretation as a proportion of variance, unlike its square. I can't think of too many uses for it and only know of one stat package that routinely reports it, SPSS. Bivariate correlation only tells you about two variables at a time (though you can use partial correlation to remove other variables).