Some more speed has been gotten with recent optimizations, bumping it up to up to 35-39% at 3MIPS in IPS clocking mode(depending on host and processor mode and features active). It's kind of hard to get faster, seeing as most time is spent performing RAM memory accesses and address verification (x86 protection mechanisms), which is kind of hard to optimize out without losing compatibility.
Edit: Said speed in full synchronization mode for the video card was the cause of the emulator's video card emulation running in the "Synchronize depending on the host" setting equivalent (with incorrect synchronization parameters), which approximates the old incorrect synchronization result on full synchronization (lowering the video card speed to make the CPU emulation reach higher speeds).