This is purely anicdotal, but the speed and performance improvement between the 030 and 040 is based more on the advances of chip integration and less an actual change in the archetecture. 030s never had a math co-processor or floating point unit embeded in it. The floating point system usually consisted of a Motorola 68882 math co-processor adjoining it on the motherboard. In the case of a full blown 040, the math co was on chip, like so many of our modern RISC processors and the Intel processors. The LC040 didn't have a math co, making it a much less desireable chip and today feels significantly more sluggish than a full 040 running at the same clock rate.
From my experience the IIfx, which had a 40Mhz 030 and a math co performed to a level aproximately equivelant to a Quadra 610 or possible 650. I still have a Quadra 950 in production for email and general AppleTalk serving. It is not the fastest or most efficient machine in the world, but it holds its own. I would not try the same tasks on a IIfx despite its beefieness.