Yours is an interesting philosophy - that you're more likely to get a 45 - 50% overclock on a pIII 700E than on an 800E. I agree with you completely on that. However, the likelihood of getting much beyond 950MHZ is probably not quite the same with either chip. Remember, these are both essentially the same device, tested to meet certain guaranteed run rates when they are tested at the factory.
Intel themselves has pretty much admitted that getting more than about a MHZ from the PIII core is not reasonably assured - remember, they pulled everything above 1GHZ from the PIII offerings.
I'm guessing (and this is purely a guess) that a device must run at it's marked speed +/- 15% reliably for it to be sold by Intel at that speed - 15% seems like a reasonable margin?
With that being the case, I'd be fairly confident that a 700MHZ device will run at 785 or even 800 - and if the yields are good in general, many will exceed that by two to three times the margin - up to maybe 933MHZ (yes, I picked this number because of the high number of successful 933 MHZ overclocks)... and occasionally even 1000 MHZ. It appears that 933 is being seen in about 70% of the devices so the core appears to be pretty much a nice solid 933MHZ device.
Maybe 20% of the devices that make it to 933 will also run at the magical 1000MHZ. Many of these are probably pulled by Intel to meet their requirements for 1000 MHZ devices. The rest of the "hot chips" get configured and sold as other speed devices - someone may be able to confirm this but I think the actual device ID is written in a writable control store on the chip - after it's been tested and it's speed determined and if they only need 700MHZ chips on "tuesday" they'll only test to 700 MHZ specs, regardless of the potential of the devices.
If the same margins are expected of an 800 MHZ device, I'd expect all 800 MHZ devices to be able to run at rated speed +/- margin. If the margin is 15%, I'd expect essentially all 800 MHZ devices would run at 920MHZ. Again, if yields are similar, there would be about 20% of the 800MHZ devices capable of 950 - 1000 MHZ.
Rather than looking at the likelihood of a percentage of overclock, I'm a lot more comfortable with the idea that a chip has proven to run reliably at a higher speed +/- margin tests will have a better chance of running at or near the tested and proven margin tested speed.
Naturally, my reasoning could be flawed but for now I'd say that a higher percentage of 800 MHZ devices are likely to run at speeds exceeding 900 MHZ than 700MHZ devices reaching the same speed. Overclockers.com's database seems to be confirming my thinking with nearly 90% of the 800's reaching an average speed in excess of 900MHZ and a somewhat lower percentage of 700's reaching the 900's.
Not too surprisingly, since they are essentially the same device, most of the "E" devices run up to about a gig maximum (850MHZ average) and up to almost 1.2 gig with extreme cooling. For the higher spec'd chips the average was a bit higher than devices spec'd lower.