Ok, I'm back. I got the motherboard working after disassembling the socket and pouring some alcohol over the contacts inside.
All the following benchmarks have been done using a Radeon 8500@9100 275/275 and 512 MB of RAM running 1:1 with the FSB at 2-2-2-5 timings. OS was a clean and slighty tweaked XP SP3 install.
A true comparison would have required a P3 1133, which I don't have. Instead, I've used a P3 1000 EB two times, at first set at the same bus speed as the Celeron, and then at the same clock speed* (*almost the same, due to choices the PLL chip offered me).
System setup:
Chipset VIA 694X/686A with 4in1 4.51v
Radeon 9100 with Catalyst 5.8
Yamaha 724 with drivers 5244
Intel PRO 100+ with drivers 8.0.47
1) Celeron at 1147 MHz (8.5 * 135)
Sandra 2002:
Super PI 1M:
3DMark 2001 SE (default settings):
Quake 3 timedemo (1024x768 32 bit colour): 94.4 fps
================================================================================
2) Pentium 3 at 1012 MHz (7.5 * 135) - same bus speed
Sandra 2002:
Super PI 1M:
3DMark 2001 SE (default settings):
Quake 3 timedemo (1024x768 32 bit colour): 103.3 fps
================================================================================
3) Pentium 3 at 1125 MHz (7.5 * 150) - same clock speed
Sandra 2002:
Super PI 1M:
https://i.imgur.com/9t9Sg43.png
3DMark 2001 SE (default settings):
Quake 3 timedemo (1024x768 32 bit colour): 113.3 fps
================================================================================
I've come to the conclusion that it's the P3 that's faster than I thought it would be and, by extension, so is the Celeron. But no matter what the lesser cache does have quite an impact.
And I guess Sandra results have to be taken cum grano salis.
Andrea