I'm trying to optimize my x86EMU emulator to reach at least 386 speed (7800 Dosbox-style cycles) with the pure interpretation core it uses(without giving up accuracy on the currently emulated hardware). So far I managed to squeeze out 20% of it's speed (So it would be equal to a 386 with only a 6.6MHz clock crystal roughly, instead of the usual 33MHz according to the Dosbox wiki). Still need to implement the missing 286&386 instructions and task switching to get that one fully running though. The instructions themselves won't be that much of a problem using the current 80186 core as a base (it's a part of the 286/386 CPU instruction set after all, and it already supports the 32-bit MMU and ModR/M addressing&protected mode). The task switching is still a bit difficult to implement atm(concerning all the security checks required during the task switch itself, including triple fault&double fault handling). Also the 32-bit interrupt handling still needs to be implemented.

I am wondering though: how did the Dosbox creators make Dosbox so fast, even using a simple interpreting core? Any stuff to watch out for when optimizing it to become fast enough?