I was messing around with an Asus motherboard from 1993 with AMI bios and a 486-66 yesterday. When the turbo was in the slow setting, the l1 & l2 cache were still enabled, but something was definitely odd about the memory access and the caches were not as effective as they should have been. The speedsys memory stairs had some odd divots that you wouldn’t expect to see if everything was working correctly. Even though there is no CPUid, most info tools still reported it as a 66mhz cpu on a 33mhz fsb.
With the l2 enabled and the turbo on slow, the speedsys benchmark was ~13.5, which would be like a 25mhz 486 with a good cache.
With the l2 enabled and the turbo on slow, the speedsys benchmark was ~15.5, which would be like a 30mhz 486 with a good cache.
With the l2 enabled and the turbo on fast, the speedsys benchmark was ~24.5 whch is pretty close to what I’d expect for a 486dx2-66 in an old motherboard that doesn’t have a write back L2 cache.
Some people wrote more about it here: 486DX2-66 Write-Back and Turbo mode
Here are some posts other people have done with some graphs:
486DX2-66 Write-Back and Turbo mode