Well, for all these landscapes/nightscapes/cityscapes images, you can also go the route of assembling...
Get Hugin, snap 4-6 pics, and you'll be rubbing all those FF noses in the dirt... Double the shots by underexposing, and you can even throw in a little HDR... It's so easy to use it really makes you think twice about upgrading.
Say you envision a photo that needs a 18mm... Make a "panoramic" of your 18mm photo with a 35mm (or a 200, but this would be an awful lot of pics!), and assemble them.
I'm currently experimenting with the Brenizer method. This guy simply used panoramic methods with portrait, so as to simulate a medium format with an APS-C sensor. With 6-10 pics taken handheld, you end up with a virtual sensor of about 7x5cm, which is simply HUGE in digital! Not even talking about the resolution here, as you can easily reach 30-40MP with full-size base jpgs.
And as the lens used keeps its properties on this virtual sensor, this brings you equivalent lenses unheard of as of today : just imagine a 50mm f/1.4 on the 645D...
Take this boring shot (taken for test purposes):
Well, it was assembled from 9 pics from my K5+FA50/1.4, and the final image was something over 40MP, with a virtual sensor the size of the 645D. But the lens is still a 50mm f/1.4, which is simply unavailable in 645 format. The equivalent APS-C lens would be something around a 21mm f/0.7...
EDIT: for some applications, you could also go the route of Super Resolution... It takes a bunch of identical photos and combines them mathematically in order to extract the sub-pixel details. It can works wonders.