I personally don't have a need for that much processing power, nor the time to write enough software to take advantage of it.I guess that you never programmed something like a small TFT Display (320x240 pixel). Such a simple element like a progressbar brings any ATMega MPU on the border of usage. Atmel's xMega's can help to solve those problems because they clocked 30Mhz but the progressbar is not the only thing that the mpu should doing. To use bigger Fonts need do define them as a hugh uint8_t array an this consumes storage space in the Flash memory. An access to that memory from a running program, consumes many cpu time. This is reasoned by the modified harvard-architecture design of the ATMega's, that the compiler not supports. GNU-CC is designed for used with a von-Neumann architecture. To get access to the flash memory, the compiler designer had created some macros that will solve it.
I am very surprised about the speed of a simple Cortex-M3 that is internaly clocked with 78Mhz. The size of flush memory is 256kbyte (STM32V103C) and the SRAM size is 48Kbyte. This chip price is cheaper than an ATMega 2560 and the tools are cheaper too. The backside of the medal is, that the architecture is more complex. I guess, that the ARM Cortex M3 / M4F (with DSP - every floating point application will love it)
will be very successful like ATMega and PIC's.
I have a couple of applications that the power of the ARM is required. I will be with you, i also know a couple of applications that does'nt need a "big gun" like an ARM. ATMega's MPU's are are very versatile in usage and is available as a DIP variant's. This are very important for test and evaluation purposes. I don't will miss them and will use them in the future too.