To me, it's a microcontroller as soon as it has GPIO integrated and intended for high-speed use (in relation to the core operating speed). Timers, serial ports, CAN, ... are just icing on the cake.

Switching between a tiny 4-bit microcontroller and a 32-bit one just affects the number of tools from my tool chest that may be applicable when trying to design the software. The smallest small may require almost everything implemented in a loop because the interrupt system is limited or the RAM limited to just the processor registers. More complex chips will allow me to make use of complex interrupt controllers and FIFO or DMA. Bigger chips may have enough RAM/ROM/clock speed that I can design part of the application to be controlled by a RTOS. The biggest chips may reach the boundary where you may use them for PC-class problems, but normally at the cost of hard (or at least fast) real-time. A PDA for example may have a microcontroller that gets used with PC-class software and with the loss of most (all?) real-time capabilities because of the choice of OS and programming models.