I hesitated adding to this thread, but I think Ecaroh (and ReBased) are having a correct analysis of Ptq's weakness in the attack portion of the sound.
As much as I like Pianoteq, I have always been critical of this point too. I think the current attack is quite correct for some repertoire (mostly classical fast virtuosic playing with far-micing), a lot of real piano recordings sound just like that, but fails in close micing and melody outlining, something more common in jazz and romantic music.
There has been recent demos of Roland's V-Piano, that although they sound slightly more artificial than Pianoteq to me, do have that closely recorded attack.
For example see this thread : http://www.pianoworld.com/forum/ubbthre...V-Pia.html
A good example amongst others: http://www.box.com/s/ho6sufuim3d334db7dpu
Roland says the attack is fully modelled, and I tend to believe them because I can always picture a real hammer in these demos. One thing that makes me believe the modelling is the blooming of the note just after the strike. This is absent in Pianoteq. I don't think a hybrid approach could produce this bloom.
For example, I combined using a simple crossfade the first 100 msec of a sample (Sampletekk's Black Grand) with a C3 (the preset) note minus the first 100 msec and although the attack portion is audibly (and visibly) slightly better, there is no such bloom in the result.
Here is the example: First the straight C3 sound then the BG, then the combination I did:
http://www.forum-pianoteq.com/uploads.p...attack.mp3
This is not meant as a glorification of the V-Piano, I think they are not in the same price league for one reason, and also the continuous part in Pianoteq is more realistic to my ear, as well as the tweakability that is far superior (yes, you can transform it in an imperfect instrument quite easily)
I'm really guessing here, but I think the generality of Pianoteq's model might be the problem. What other software can produce along with piano all those other percussion and bell instruments with such realism. I think struck percussions and plucked instruments as harpsichords are rendered quite well with a common general approach that probably automatically extracts the attack portion and models the continuous sound. Piano attack, though, is in a class by itself, since the attack influences the continuous sound in such a complex way.
Some instruments, like organ, can be perfectly rendered with samples since the speaking portion of the pipe would be hard to model, but easy to capture by sampling and the steady portion can be looped without problem. I think modelling is the right approach for piano since the continuous state is so complex, but the very important attack portion needs to be modelled too for completely satisfactory results.