@Kwag and all with programing skills
One crazy idea (I think crazy enough to work):

I noticed that in the output tab you can choose to save a ffstats file. If you are making a second encoding of the same source you can choose to use the previously created ffstats file, that way you can do a second pass encoding.
I opened in a text viewer such a file and I noticed that is in plain text. I think that you or somebody with programing skills and more understanding than me of how things work can understand what is in that file and how it can be modified to work the way we want.
My idea: it could be great if we could feed FFVFW first with a sampled file (made with sampler.dll), modify the obtained ffstats by extrapolating the values and then using the resulted file for final encoding.

What do you think?

you already can do that... there is an xvid tool that can access ancd change the file...

but also ffvfw can do so (thats what he does in the 2nd pass )

You didn't pay attention. The ffstats file that I'm talking about is diferent of the stats file ffvfw is making during first pass encoding. That one is unreadable directly by any human. The one I'm talking about is made during CBR encoding and actually you can see in it how encoder works. But this one is unreadable with that Xvid tool.

oooooops
To extrapolating the sample we only need its size and length (as done with CCE, TMPGEnc and so on)

But AFAIK we couldn't do much with I,P and B Framevalues because they are only the results for the sample we encoded... the other part of the video doesn't have anything to do with this.

(But only AFAIK... mybe it does )

Not crazy at all
You should see some of my crazy posts The problem is that these statistics are for useable if we have control of the encoder. For example, even though we can have a footage analysis with the stats, we can't apply them to TMPEG (or other encoders) because we can't determine what the compresibility of the encoded mpeg will be. Every encoder will have different internal algorithms, so we can't tell what the output will be.
Only if we have full control of the encoder quantization, etc., then we could apply the stats for a post-process.
So the stats created by ffvfw are really usefull for it's internal second pass, when it finally works!, because right now it bombs on the second pass.

-kwag