SM3.0 is PS3.0 + VS3.0..
PS2.0 can emulate PS3.0 using more passes with bigger performance cost ,and in some cases inefficient workarounds.
(with the exception of the limitations in PS intructions counts ,which you cant workaround . limits are limits.)
VS2.0 can't emulate VS3.0..
it cant do GEometry Intancing ,True Displacement mapping neither FLow control. to emulate DM in software ,with previous VS2.0 hardware "is possible" but throught workaround (hacks) using the CPU .. that doesnt look as good as the real thing DM.
the things are more interesting when both are working together. that is SM3.0 . so if something is done very quickly under Sm3.0 and its very painfull to do (but possible) under SM.2.0 and cost too much performance what do you think developers will do in the end?
"True" HDR lighting (64bit FPblending/filtering),is possible only in the NV40.. (thanks to another feature that is not part of SM3.0 yet) very easily. in the ATI cards nothing like that exist. so HDR lighting is done with many limitations to the point that now many incoming games will use HDR lighting but only in the NV40 hardware.(SplintelCell3/Unreal3 and others). Developers like valve$$ will not have a choice$$ to do HDR in the limited situations that ATI R3xx/R4xx hardware can do it ,using workarounds and in the hard way.
There is no rocket science here , SM3.0 is simply SM2.0 without its limitations. DEvelopers will always have preferences for the quickest and more efficient ways to implemet prettier graphics in their games.
this is common sense .they will not ignore SM2.0 hardware.. because the biggest base is there.. but the best looking effects will be done in SM3.0 at higher precision. if they want to deal with the limitations in PS2.0 hardware. if up to them.
so question is not more is there are diferences ,but when you will see them?. and by what i have seen there will be a few good games with some support SM3.0 before the end of year.