those opening "wtf" sequences might be there as filler space; harmless instructions with a known pattern where you can come back later and insert different instructions. Most people use NOPs for that but perhaps they wanted a different signature or needed 3 separate, differentiated patch points at entry. Or maybe they wanted to help sell more 8087 chips.

Anybody recall if there was a notable performance difference between Borland's FP emulation lib and M$, then? My habit at the time was to religiously avoid all floats, to the point of shipping a home made arbitrary precision BCD math library. It was no faster than anything else but it gave the same results for the same inputs, every time on every machine.