I tend to say you're right. As far as I know, there was first the 486SX-20MHz that Intel introduced for marketing reasons in times Am386DX/40MHz was already on the market. The idea of having a processor which was names a 486 (I do not speak here about Cyrix 486SLC-33MHz ... I believe this is actually a 386) was overwhelming but people didn't bought large quantities of this processor.
Ok, let's talk about the Cyrix (and a few others) CPUs. There are a few confusing part numbers, but "SLC" means that the CPU would have been a replacement for a 386SX. There were later versions that doubled the clock speed internally (for both 386SX & 386DX replacements), and had 1Kb L1 cache or more (activated by software). This is right before the time Cyrix started working with Texas Instruments & IBM (mainly for legal challenges from Intel, which TI & IBM had cross-licensing agreements with).
IBM was allowed to modify the Intel 386SX code mask & produced a few CPUs that were able to be put in the limited 24-bit memory addressing (16Mb RAM maximum) and 16-bit data platforms for a little more processing power. The IBM 386SLC was a single-speed CPU, but added 8Kb L1 cache (and interestingly ran all 486SX instructions too). Later 486SLC2 & 486SLC3 doubled and tripled the clock rate internally, and add a 16Kb L1 cache.
Recently I benchmarked an Intel 386SX-20, & IBM 386SLC-20, 486SLC2-40, and 486SLC3-60 in an identical environment (actually the same system, because the upgrades are in the form of being a daughtercard) with Landmark 2.0 (a common DOS version out there). Some platforms even have the 486SLC2 at 50MHz & the 486SLC3 at 75 or even 100MHz! Here are the scores (results are given in the MHz speed needed for a theoretical 286 CPU to equal the processing power):
386SX-20: 20.68
386SLC-20: 45.92
486SLC2-40: 91.68
486SLC3-60: 136.87
Then I remember was the 486-20MHz (without SX) had the 8Kb of cahce and essentially was a revised version of a fast 386DX but including ca co-processor (FPU). I never remember seeing one of these in the flesh.
All Intel 486 CPUs started at 8Kb of Write-Through cache, with it increasing to 16Kb for the DX4 line. A few CPUs at the end of the DX2 & DX4 series had the ability to set the L1 cache to Write-Back mode. Look over
www.gilanet.com/David/Intel486.htm for more information.
386DX-33MHz was as far as I remember a very appreciated processor of the time.
Which is amazing for the amount of bugs the Intel version had. Intel didn't kill all of them until the F-step, which was some years later after production started. By then the 386DX market was pretty dead anyway & I've never seen an F-Step (visibly diffent label is the only way to tell).
But about the 486SX-25 I've got nothing but bad words. It's in essence a 486DX with the co-processor disabled. Why Intel chose to build the co-processor and hardware-disable it in the chip, I don't know.
Only the early 486SX were made this way (and allegedly at the start when the FPU section failed). The 486DX were made first, the 486SX was a later "budget" CPU (just like the 386DX & 386SX). 486 CPUs would have to be tested before the packaging was complete, because their CPUID tells which model they are.
Indeed, for stability at least heatsink / passive cooling is required for a 486SX-25 to run at 33MHz. You are perfectly right but if not put to hard work, it'll do without a heatsing although life span shortens...
I do it just to avoid cooking things...
This classement is done through my experience that might just not be that great si I expect some critics! I could learn some things now, as I missed the right time...
But came to the right forum!...
PS: Indeed I've heard (but never seen) about a Harris 286-25MHz processor outperforming a 386SX-25MHz...
Just look at the MHz value the 286 has to be at to place equal to my 386SX score above...