Following up on what Paul said,
We can run the emulation on a Raspberry Pi pretty easily without any system lag. We're using a fairly beefy PC setup right now (compared to other single board computers) for rapid iteration of different ideas and such. Once the tech stack is finalized and we've got a pretty good baseline down, we'll start wrenching down on optimizations for the hardware set and target compute cost that we're trying to achieve.
We went through quite a few hoops getting TNA to run as well as it does on the hardware system it has. However, what ended up happening is we built something that is quite resilient, cost effective and has a very long shelf life. For example, the OS and underlying code has no direct ties to the hardware, so we could port it all over to something else if needed, or someone in the community could do it if we're not around anymore in the far distant future.
Those boards give us plenty of headroom, but you make a great point... Many times it comes down to how good the programmers and architects are. For example, you need to leverage as much hardware acceleration as you possibly can, and you can't be doing crazy things with asset loading when your slowest peripheral is the disk. It's a constant balance, but it's quite fun!
-- Jimmy