Apologies for bumping this old topic, but I found very little discussion of Dietmar Meschede's tools in relation to DOSBox performance here or anywhere else.
I got the latest versions of both S3 VBE/Core 2.0 and S3 Speed Up from the Wayback Machine copy of the author's site.
I've ran some tests with resource-demanding FPS games, most thoroughly with Duke Nukem 3D and Descent (both shareware versions for simplicity's sake). I have used the current vanilla DOSBox build v0.74-3 and the old build by Gulikoza with the FPS counter, like the OP here as it seems.
I kept most DOSBox settings in both instances at default, except for cycles which I capped with the limit function at 26800 (~486*) and 77000 (~Pentium*). Aspect ratio correction is on.
* -- as per the table on this page
With mode 13h speed-up (VGA+), so far I'm only noticing a slight increase in Gulikoza's FPS ticks (of up to about +4) in Descent when running at max 77000 cycles (for testing, I recorded a short demo of the first level and run it at various settings). At 26800 cycle limit there seems to be no tangible framerate improvement with S3 Speed Up.
I also tested the VGA+ mode with the demo of The Elder Scrolls: Daggerfall (which is known for very uneven frame rate), again at 26800 cycles there was no notable improvement as far as Gulikoza's FPS counter goes.
With Duke3D, there's additionally a built-in frame counter that can be turned on with the DNRATE cheat code. Also things get a bit more complicated here because for the 320x200 resolution, Duke can run in mode 13h, VESA mode 150h, and by editing the configuration file, you can also get an "S3 optimized" mode which indeed seems to be slightly faster in DOSBox even without S3 VBE/Core 2.0 and Speed Up.
For the VESA speed up (VESA+) of Duke's regular VESA 320x200 mode, the slight increase is again only noticeable when running at max 77000 cycles. Gulikoza's counter does not go above 70 FPS, but seems to stay near or at that mark more consistently with S3 Speed Up on (here I recorded a demo as well). The results are largely similar with mode 13h, i.e. no notable performance increase with the Speed Up tool.
Maybe I'm doing something wrong here? Perhaps I should've loaded an external VESA driver first?
UPD: I realised that DN3D is apparently using linear framebuffer VESA modes while S3 Speed Up is for banked mode. I grabbed the NOLFB utility from Ken Silverman's Build page, hopefully it will make a difference this time.