But it's not MCU itself that's the problem, it's the peripherals!
I think you're in danger of contradicting yourself. If an 8-bit CPU has peripherals that do what I want, why should I convert to a 32bit CPU with brand-new overly complex peripherals "documented" by providing some bloated vendor-provided (and controlled, and
maybe supported) library?
There is interesting stuff happening along these lines in the Arduino-compatible space. There are now a couple of 32-bit Arduino equivalents (Maple, using ARM, chipKit using PIC32, Firebird on Coldfire.) They have varying degrees of SW compatibilities, and about the same prices as an (8bit AVR based) Arduino. They're not gaining wild acceptance; it seems that most projects fit in the 8bit AVR just fine. Peripheral libraries seem to be a problem; I know chipKit has incompatibilities introduced by using the Microchip-provided peripheral libraries, and IIRC Maple tried to use the CMSIS libraries and had problems. It really is NOT a "feature" when accessing your peripherals becomes so complex that you have to rely on someone else's code. (for instance, the core chipKit code ends up being bigger than most Arduino AVR sketches. Not that that really matters, given that it has so much additional code space.) (The AVR Arduino libraries are written on top of bare metal.)
That said, there's an increasingly broad range of applications (eg networking) where the solution for 8bit CPUs is an external network processor with capabilities far exceeding the CPU (the first ethernet interface for Arduino used a Lantronix XPort, which is an x86 class CPU almost capable of running linux (newer XPorts DO run linux!) That doesn't make sense either.
In some sense Arduino exists because desktop CPUs became "too powerful" and disconnected from the physical world that people wanted to access; 32bit microcontrollers are in danger of going the same way. Sure, I can get a $100 board with lots of MIPS, USB and Ethernet peripherals, but the $10 of a stripped-down AVR board (home assembled) isn't really there yet.
It's also hard to tell how much things really cost; too many "development boards" are subsidized by manufacturers. You can get a mass-produced ARM linux system (wireless router, pogoplug, etc) for $50, but how much will it cost to get 100 or 1000 of a board based on a 32bit cpu? (and yes, a lot of "products" are built in that sort of quantity.)