I just don't think it makes sense to code GNOME Shell to the limitations
of other pieces of the software stack. The effort to fix the other
pieces of the stack - to create free software ways of doing thin clients
with a composited snazzy desktop - is going to be comparable or less
then the extra effort we'd need to put into GNOME Shell, and the end
result is much better.
And the same applies to virtualized desktops, but more so. Don't put the
effort into avoiding 3D. Put the effort into making 3D work.