Here is the issue...from my testing the encoding (most apps) dont care about the cache that much, hence why the celery does well...probably comparable to the p4 of equal speed...On the other hand I thought the celeries dont have HT....In that case an HT app in alot of encoding apps can give 10-20% speed boost...I dont know about your app cause you have not listed it. If the app you run is HT aware Quit being so cheap and get a 2.4c or 2.8c and boost it to 3.3-3.5ghz range and it should destroy that cel D.
Now from my testing and I had a 2.4c@3.5ghz (extremely oc'd bus) and now this 3000+@2.66ghz I see that in most instances I am equal in these same HT aware apps to about 3.3ghz of my P4 or about 7% slower......At 2.0ghz like your cpu that would be a 30% slower then my curent mark, and my test show they increase pretty linear with ocing So your claims of 1/3 slower could be legit.
My guess would be that a 2.66ghz like mine could come close to equal that Cel D, but would still lose in most all apps versus a 3.5ghz p4 (especially if oc'd a bit).
I use TMPGenc and it can do SVCD and it is extremely HT aware..couple this with a cpu with HT and you can have gains of 19-22% over same speed chip without HT....
I do less and less encoding now, so for me a bit slower in this area didn't concern me. I am faster in most all of my other apps I use more frequently....PLus I enjoy the quieter and cooler running chip. I also got the sckt 939 to poise myself for dual core chips without having to upgrade my platform over and over again, and to start messing with 64bit cad apps. It wasn't much of an upgrade...more lateral then anything, but as an ocer I was bored and need something new to play with. I fully expect to get a venice core and push for 3ghz here shortly.