Billy Mitchell just happened to use older MAME versions without those improvements, even during his 1.062M run that was performed several years after MAME 0.127 shipped.

An actual arcade board renders the levels with a "sliding door" effect, while MAME builds entire chunks of a level at once and then displays them as a complete screen buffer (depending on the version).

It's just evidence Mitchell's score is not what he says it is.

If you agree with the premise that an emulator should strive to... emulate the real thing as closely as possible, then it is most certainly a bug.

http://mamedev.org/about.html

The creators of MAME are mainly interested in rigorously simulating certain kinds of hardware which happen to play games, just as a classic car restorer may rarely or never actually drive his meticulous reconstructions.

For many other emulator devs, and almost all emulator users, the point of emulators is to play games. That may not be your area of interest, but it's rather silly to say that they're just wrong.

Besides that, the real console hardware could vary more between revisions than the difference between a console and an accurate emulator, so you also have to ask whose real thing is being emulated? Even with Nintendo's attention to detail on lot check, some real Game Boys won't play Prehistorik Man. There are some NES accuracy tests out there that require making your emulator behave like one specific console owned by blargg.

This is exactly what BSNES set out to do. I think it's called Higan now > https://arstechnica.com/gaming/2011/08/accuracy-takes-power-...