dirkmirk, I agree with you on the systems that should use an ISA VGA card.
About the benchmarks: My strategy when comparing ISA VGA cards is to measure linear throughput when sending data (writing) to the card. Most fast-paced games calculate the image in system RAM and then copy it to the card so this would be the relevant metric.
The tool I prefer for this purpose ist c't magazine's ctcm http://www.heise.de/download/ctcm-111825.html ftp://ftp.heise.de/pub/ct/ctsi/ctcm17a.zip (Unfortunately it doesn't run in virtual mode.)
The documentation is in German but with ctcm7 /Vid you can measure video throughput. You can use /C=US to get english output. This works also for the integrated help screen: ctcm7 /? /C=US.
I just checked out Speedsys. The number after "VESA memory" given in KB/s should represent the same value. I just can't find an option to set the video mode used, I know that this can make a big difference on PCI/AGP cards.
When I first upgraded from a 486 to Socket 7 (K5 PR133) before I bought a PCI video card (S3 Trio64V+, good choice in retrospect) I noticed that the actual transfer rate went down, ie. it was slower than on the 386 even though actual game performance was of course faster. This is why I think 3Dbench or game benchmarks are not very well suited for comparing the raw performance of different video cards when they are not tested in an identical system. Quake would be CPU limited on any 486, according to the mentioned benchmark a sufficiently fast Athlon is able to push 36,2 fps on the fastest ISA cards which is more than any 486 can do under circumstances (except maybe overclocking to 500 MHz using liquid nitrogen 😉 ).