I'm guessing that the division which reports 2666.7 (or 500 ms) doesn't have proper accuracy, which is why it's coming up as such an exact number. Have you figured that out?
It's interesting to see that Dhrystones measured in Speedometer 4 and in NetBSD on the same machine come pretty close. I'll have to download this program and try it, too, for comparison.
Dhrystones in Mac OS 8.1 on Q610 (50 MHz m68040) in Speedometer 4:
35888
Same machine in NetBSD 5:
36764.7
For comparison, it seems that the m68030 in the SE/30 beats out the processor in the VAXstation 4000/30 (VLC), which is the slowest computer I own which runs NetBSD. It only gets 2164 Dhrystones a second.