No, you are not raining on my parade. Just making a suggestion based on a different view and different requirements.

Try to produce 10k units with four extra chips - even at $1 each, that would amount to $40k. Multiply by time, and that will represent significant sums.

These chips would consume board space. Board space represents money and affects the size and shape of an enclosure.

They would consume power, unless they can be individually turned on/off, or at least held in reset. The extra power consumed by the higher CPU load of a host processor running virtual machines would end the same moment the virtual machine is stopped.

They may possibly need their own reset circuits, unless the master processor dedicates a processor pin/slave for the reset.

Being slaves, the master would need to constantly poll them, for information they want to send back.

The chip would also look like an embedded chip, possibly require the user to know embedded programming.

I/O sharing would not be an option, since a slave processor would not have the number of pins to share any significant number of signals - and even if they could, it would take two extra layers in the PCB.

The master processor would probably need to allocate 11 processor pins (MISO, MOSI, CLK, and 4x reset and 4x slave-select), unless the signals are multiplexed or serialized.

With emulation, existing hardware can be used for at least evaluating the concept. Test results could suggest an upgrade to a master processor with more memory or that the master should be run at a higher clock frequency. But chances are that an existing hardware can be used as-is.

In the end, a solution with separate slave processors is not an alternative. The only available options are:
1) define my own virtual machine and create my own compiler for programming it.
2) emulate an existing processor and make use of an existing compiler.

Alternative 1 is conceptually tested, but with a rather simplistic language. I would prefer a full-grown standard language, but without having to write my own Pascal or C compiler. It normally takes longer to write a reasonable compiler than it takes to write a processor emulator. And the probability is good that there already exists a suitable emulator I can use. But the emulator must have a limited RAM footprint and I can't link with GPL code.