Really nice bench… there is one thing at the conclusions, it would worth to check, which technologies were available at the specified year. Eg these highend scores based on very late 486 boards, setups, cpus - so if you want to compare, you should mention the typical use cases, which was really available. Eg:
* the p100 was available since 94, but practically just in a very low amount, so the typical setup was the p90. the p100 become more popular in 95, when the higher clocked p120/p133 was also on the market
* the typical l2 cache was until 95 the async wb cache in the pentium (socket5 boards), the pb gives +10%, but became popular in 96
* the pre-triton chipsets, boards was also slower, so if compare a p100 in a 94 setup to a 96 setup, it was a significant difference
* no proper l1 wb support before mid 95 for socket3 (practically you could buy at late 95, early 96 a mainboard with wb support) - many times you could enable it, but the realization was buggy, or simply the cpu was in wt mode. A cache bench shows the difference.
* intel released the wb based dx4 just at the very late of 95, no wb cpus on the market
* check the chipset availability - the tested umc board was probably the direct competitor of the pb caches pentiums, it was not “2 years newer”
* the support of the 5x86 in 1995 was very limited, and the cpu ran slower in these boards/bioses
* the pentium pod released just in 95
So i would compare period correct 94/95/96 etc cpus, mainboards, bioses, anyway you will compare 2-3 years old technologies, and something sounded way better in 94, than in 96, or they perform differently in a period correct setup.
* The intel dx4 wt was available in 94, the amd in 95, the 5x86 with the proper mainboards just in 96, when mainstream was much more the p100-133 with pb cache tritons.
* If you test an intel dx4 and the 5x86 in a board from 95, they will give very similar performance in fps tests on average, but it can be +-10% performance difference in the tests, etc