That's an interesting topic to discuss. On one hand, mt32emu engine consists of several "truly DSP" parts, like analogue path filters. On the other hand, those parts take not that much CPU load. Still, we can't just say that it isn't audio-specific; this rather depends on the synth model. I suppose, the float LA32 emu engine is kinda more audio-specific than the more accurate integer LA32 emu model, because the latter heavily uses those bitwise operations and LUTs.
There is one more approach to consider if the use case permits it. Instead of relying on the audio DSP hardware, we might use GPU/APU to do the heaviest bits of the synth engine work. That seems to be a good choice for configurations that involve many partials to emulate, but I don't get the use case atm. What I did try is I looked at adding support for OpenMP and OpenCL/Vulkan for it, though I didn't try too hard since I have 0 performance issues with my hardware (including the phones and the tablet).