Since I wrote several significant applications in assembly on the ADSP-21xx using it's bits and pieces hardware to do that and, of course, this required care and attention to numerics, I kind of agree. None of what I wrote was general purpose FP, since I was able to fashion each complex calculation exactly as required, retaining exactly the desired precision and guard bits I wanted all along the way. Quite fast, that way.
Most people using floating point have no idea of the risks. Doing it at the level I had to, you could NOT know about the risks.
Also, FP hardware simply sucks power and die space like no one's business. Increases die cost and increases the overall BOM, power dissipation issues, and likely board space as well. Yeah, it's for weenies.
:)
Jon