AMD stumble out of the gate?

Just recently samples of 65nm AMD Athlon64 X2's began to make the rounds and get reviewed in all manner of places, and something very peculiar became apparent, the newer chips are slower than the, on the surface at least, functionally identical 90 nanometer parts.

Further digging was done, and it appears that the 65nm chips are not in fact a straight die shrink, they made a few little tweaks here and there, the most notable of which seems to be an increase in the latency of the level 2 cache, from 12 cycles to 20 cycles, which I'm sure you'll agree is a whoppingly massive increase... and if you don't agree, then your face;

As although the Athlon 64 has always weathered cache misses well, on account of its on die memory controller giving low physical latency to main memory (not to be confused with the cunning memory disambiguation scheme Intel's Core 2 Duo processors use, which gives low effective or as I informally refer to it, 'tactical' latency.) having 66% more of a wait to get data from L2 isn't exactly a good design decision, at least from a performance perspective.

AMD have stated that the L2 latency increase is to give them the ability to release a chip later with significantly more L2 cache, which is all well and good, except it doesn't pass muster as an excuse, they must be smoking some very good material they sourced in Amsterdam if they think that convinces anyone. and here's why, a processor with more cache on it would need to be respun at the silicon level anyway... simply to add the cache to the mask, and so any pertinent adjustments to timings and layout would be best made there and then, rather than crippling the part ahead of time for no gain, and some loss.

I consulted with a few people, and two possible real scenarios arise :

One)
AMD have done it as pure yield+binsplits improvement measure, simply trying to get as many working high clocking dies off a wafer as possible to stem the bleeding until they can get something that directly competes with Core 2 Duo on performance into the marketplace.

Two)
AMD have a real problem with their 65nm process and had to slow down the L2 to avoid a hotspot issue on the die or similar, whilst a hotspot in sram would be pretty remarkable, there's also all the cache control logic, slowing the cache down may have been to give that a breather.

In short, if you're looking to buy a machine now, and aren't on an ultra-tight budget get a Core 2 Duo, AMD's lineup simply isn't compelling beyond the very bottom end, and this latest slip in raw performance for your money certainly isn't helping matters.