I expect IPC to be close to irrelevant. (...)
Something is slowing down the Linux machine considerably.The time difference is, thus, 146551653 to 95262209, for a factor of about 1.538. Lower than the 1.543 preliminary figure, below your 1.6 but not all that much below it.
But even if the Core i7 was 100% throttled and never turbo clocked to 3.5 GHz (which would make it 2.18 times faster, just by clock, if the rest of the cores were idle), the clock difference would be 1.625 difference.
The Pentium 4 had horrible IPC, and even worse ability to branch without huge penalty, so I don't believe that there are any real workloads which would show a Pentium 4 run at the same speed for the same clock rate as a Core i7 6700HQ.
You mentioned that the overhead of setting up stack trampolines on newer NetBSD, and I have to wonder whether there's some overhead issue on whichever version of Linux you're running.
I'm really curious about this. Which versions of NetBSD and Linux are you running? Perhaps it'd add a good data point to run an older NetBSD on a newer processor to compare.
I don't have a real KA630 set up to try this on, so I don't know whether it's a bug in the emulator or a bug in the kernel I'm booting. I did instruction traces and the only differences are ignorable right up to the point at which it finds the message buffer already set up; I haven't yet followed my traces past that point.
What other systems will that version 1.4T run on? I can try on a VAXStation 4000/60, if you like.
John