The critics are right

Thats rather vague. What makes you say KDE 4.0 is overrun by "Vista-loving usability nuts"?
The oxygen style theme, maybe? The icons and look/functionality of the applications certainly
wouldn't be. A more realistic complaint would be that KDE 4.0 was gnomified, with lots and
lots of features 'removed'. But even that would be disingenuous because the feature
regressions are mostly/entirely temporary, as the features were only lost to simplify the
porting to Qt/KDE 4.0, or haven't been implemented as it's an entirely new application (like
is the case for Plasma).

KDE 4.0 is definitely not ready for prime time, and many of the developers were trying to
emphasise that prior to the release of 4.0 (though there was some confusion with KDE 4 vs KDE
4.0 and other things that lead to mixed messages and pain). KDE 4.x has already improved in
lots of ways over the 3 series, looking at the KDE Games for a great example of this (I *love*
KSudoku in KDE 4... can't stand any other Sudoku program). It'll take more time for all the
applications to get to the same point as they were in 3 (but mind you, most applications have
already passed where they were in 3 in at least some ways), but they are working on it. 4.1
pre-releases are also a *huge* improvement over 4.0, Plasma has a lot more features
(Folderview is FAR better than the previous plasmoid icon method, and in many use cases far
better than KDesktop's method... in 4.2 it'll fill the last holes as well).

But what in your opinion is the direction in which KDE 4 is headed ('wrong one' isn't a good
answer ;)? Plasma's problems mostly have to do with it still being young (Kicker and KDesktop
used to crash all the time, even several releases into KDE 3 series). Oxygen can be completely
replaced with other themes just like the themes from KDE 3. Or is this all just about the
Cashew?

I've been developing an application using KDElibs 4 and find it to be a VERY nice experience,
far better than using Winforms on .Net (layout was a pita for one), or GTK (the documentation
didn't feel much different than reading a sparsely commented header file), as well as being a
nice improvement over KDE3/Qt3 (lots of little improvements that just helped with the QoL).