Now that I have moved beyond Special Edition and have cautiously bought a few strategic full libraries in each category, I'm not complaining as much about how much work it takes to get musical phrasing out of VSL :-). There really are an amazing number of patches and matrices that allow one to minimise part-splitting.
Nevertheless, this is for tracking MIDI; not for live playing. And although I don't have to do much MIDI tweaking anymore to get very usable results from VSL, I would still like to see it set up better for live playing and live tracking. It's not that I intend to use a computer at gigs (I don't); it's just that one generally is more successful using the final target sounds while doing the original tracking (due to phrasing, etc.).
This is more of an issue for me currently with the woodwinds and brass, as I reluctantly parted with my physically modeled Yamaha VL70m last week as I concluded that while it is great for live playing, the final results are just not of professional caliber (for the most part). The big dilemma though, is that I am unable to successfully map very much of the Yamaha WX5's MIDI controller data and other MIDI messaging, to VSL parameters in a way that doesn't have unintended side effects.
Fortunately, many of the VSL patches and matrices are able to infer musical expression and phrasing as well as articulation-switching, through clever cross-sampling and good programming. Some of these are the speed-oriented patches and matrices, which really do not have an equivalent in the MIDI world. In many ways, they compensate for not having appropriate actions to map to aftertouch, MIDI CC #3 (Breath Control), and less common MIDI parameters that most wind controllers can generate (for growl, etc.).
As I have only begun working in earnest with the full string libraries this past week, after first focusing on the full woodwind and brass libraries, I have not yet reached the same level of proficiency with them, but am quickly finding that I am able to get more realistic expression, even with only one patch or matrix per MIDI part, than even with the best libraries and modules out there (I consider Kurzweil to be the Gold Standard when it comes to musically useful programming; bearing in mind that workstations/ROMplers are seriously compressed).
The most critical problem I see in the strings libraries is that the portamento speed cannot really be adjusted. This has been discussed before, and I understand the reasons and that portamenti are directly sampled vs. pieced together, but I still think it would be better to have some flexible portamento than not at all. Sure, it will not sound as realistic as other parts, but certainly moreso than using an inappropriate portamento speed (as is currently the case). As a result, I bury those parts in the mix currently, even if they are meant to be key phrase accents.
I have worked with a bunch of the other libraries, and don't find most of them remotely usable in comparison to VSL (e.g. the entry-level versions of EWQLSO, MSI, and any hardware-based solutions). The one exception was Miroslav Philharmonik, which I also sold a few weeks back (I had held onto it primarily for its brilliant chorus, but am satisfied for now to just stick with sopranos via VSL). The Breath Control support for brass and woodwinds was quite welcome, and the strings were easy to use, but the detail is lacking in comparison to newer material such as VSL. It is too bad they did not take that library to the next level, as it really represented the first real stab at musical realism in a sample library.
Sample Modeling, and related technologies, show great promise for certain instruments, but I'm not convinced they'll have much succeess with strings. I still can't get usable results out of the two Garritan products (Stradivari Violin, and Gofriller Cello), but it could be due to my computer specs. Sample Modeling may improve upon those libraries and re-release them, but I wouldn't hold my breath, nor assume the technology will be applicable to ensembles vs. solo instruments.
I continue to try updates to Synful Orchestra, but find it falls down quickly once applied to faster articulated passages such as in Mozart. A brilliant effort, but I'm not convinced it will bear fruit when it comes to serious orchestral renditions of challenging works.
In my view, VSL is where it's at, and the future lies in additional programming with the existing sample set. Perhaps this is doable from the end user point of view, but some of it requires platform support (such as allowing more than two simultaneous parameter switches, and more ways to piece together multiple articulations so that assigning more MIDI controllers and MIDI parameters can produce useful results). Range-limiting could also be helpful (though this is partially doable via the MIDI vector graph, more control points would be needed).
Sorry if this sounds rambly; I don't post often here, so may have tied too many topics into one response :-(.