Microchip's bread and butter is still cheap 8-bit controllers like the PIC10/12/14/16/18 families. Many devices don't need a 32-bit core, so wasting a few cents on each unit by specifying one drives up cost at high volume.
While the core of a 32-bit controller doesn't take up much die area at smaller process nodes, core size has never really been the limiting factor with MCUs. Look at any MCU die - the flash and RAM both dominate in size over the logic. Going to a 32 bit controller generally means that your instructions are going to be wider, so your program space is going to take up more flash. You will also tend to use more RAM unless you are judicious about word size, and a context switch on an RTOS will cost more memory since the context contains much wider registers.
Another pain point with the 32-bit MCUs is that ARM currently dominates the market, and their licensing fees add a few cents to the cost of each part - which can be a killer at volume. There have been attempts to break ARM's monopoly (most notably Microchip's MIPS-based PIC32 series) but it seems like ARM will continue to dominate this space in the years to come.
As a final note, SiLabs recently introduced an ultra-low power microcontroller line based on the venerable 8-bit 8051 core, so clearly they think that 8-bit MCUs have a role to play in the years to come.
And of course they still make tons of them, it's embedded, you need to supply these chips for another 10 years for old designs.
But you shouldn't use an 8-bit uC in a new design.