I think in order to keep the development continuous and the hardware and software in sync we need a platform that isn't constantly aging out. If we had a more modular approach I think we could have a more enduring platform.
CPU - single board computer such as the Raspberry Pi 2 gives lots of power for a low price. There are lots of of other single board computers with lesser (cost and functionality) with adequate capability.
OS - I think an RTOS based on Linux makes the most sense for a stand alone ECU. That will allow for a cross platform so when your CPU is out dated or a cheaper more cost effective model comes out its simple to implement the new hardware. We also have tuning software already developed for that OS.
ECU Module - this would be the module that controls spark output with the appropriate drivers such as IGBT, etc, integrated on board MAP sensor and buffered inputs. This could be similar to the MegaSquirt V3.0 hardware on the input and output side but I would suggest having output drivers to run eight coils and eight injectors which would cover the vast majority of applications right up to full sequential fuel and spark. One of the problems would be the non standardized footprint for modules (shields, hats, etc). It would probably be best to design the ECU module so that the complete module solders onto the appropriate shield or hat, etc adapter board to fit the footprint of the prefered CPU.
We already have the vast majority of the difficult pieces in place. Tuner Studio and the MS2 firmware has already had many years of development and implementation. With the horsepower available on a modern SBC (thats single board computer not small block Chevy

With the advent of proprietary firmware with MS3 the open source end of MegaSquirt seems to have died. I think this modular design approach would be a great way to bring MegaSquirt (or some new form of open source EFI system) back into the open source community in a way that will promote it's growth and keep its cost down.