For balance, programming the Motorola DSP56001 (not to be mixed up with the AT&T 3120) will probably melt your brain unless you take great care, so I'd rather stick with the '040 FPU, given the option 
We covered the DSP56001 in more detail than I found comfortable while I was at unversity, in the lecture on computer architecture and parallel data processing.
If I remember correctly, the DSP56001 did not see much use in the "normal" NeXTSTEP operating system and application software context. One notable exception was the Mandelbrot set renderer application which demonstrated how much faster the DSP56001 floating point operations would be than the '040 FPU's capabilities. However, the GCC version which shipped with system only supported the FPU, but not the DSP56001...
First, my mistake, it was 3210. Which wasn't built in to many computers.
Being the first 32 bit FP DSP, it has obviously been outclassed by later releases. It was noted at the time that Macs could not natively support it without using Unix (only Quadras aimed at FMV editing had them built in).
What made them popular for some applications was their robustness - 2,000V was what they were rated at. Also, comparing the datasheets from the time, it was approved for 4 different government/military applications. Other DSPs might be approved for one, rarely 2 applications. Timex DSPS generally die at 10 V or so.
Currently I'm studying Aries DSP architecture - awesome ideas spookily close to the Amiga.