The first is a message board posting on Ace's Hardware that makes some observations about the relationship between a T-Bird's multiplier and the configuration of some shorting bars on top of the chip's packaging.
It seems that the pattern of one such set of shorting bars is different for each multiplier. In fact, the settings match a table of multiplier settings in the Thunderbird's technical documentation. Another set of bars corresponds to a table of voltage settings. Based on this, it's theoretically possible to alter the multiplier and voltage on a T-Bird using a hobby knife and a #2 pencil.
A brief explanation as to how the things likely work: the four multiplier bars are all connected at the beginning of the manufacturing process. When it's time to set the multiplier, laser etching is used to sever some of the connections, establishing a binary sequence that corresponds to a specific multiplier. The message at Ace's Hardware explains it fairly well; if you look at the close-up pictures that the author links to, you can see how the pattern of the shorting bars corresponds to the multiplier table in the AMD technical documentation (page 62).
Of course, this is still speculation, although very compelling speculation; it hasn't been tested or proven yet. But take a look at the other link JC's has for us: a press release from motherboard manufacturer Legend QDI. They've announced a new KT133 motherboard with which "the multiplier of the processor can be adjusted by a jumper on the mainboard. . . ." If this isn't vaporware, the Thunderbird could very well be an overclocker's paradise.
While at some level this looks like AMD has foolishly chosen an "easily cracked" multiplier lock, I don't think that's a fair assumption to make. We're talking about the company that took the performance crown away from Intel. Look at the capabilities of the Athlon core, and tell me that AMD couldn't make a multiplier lock that's just as difficult to crack as the PIII's.
No, I think something else is at work here. I think that while Intel spouts the party line of "We love overclockers, it's those darn remarkers we hate," AMD actually means it. They created the original Athlon with a multiplier lock, but not only was it very easy to figure out and manipulate (via soldering/desoldering resistors), they stuck an extra connector on the top of the PCB in case you didn't like to solder!
Recent events in Australia has proven that AMD will pursue remarkers at every opportunity. Nonetheless, the original Athlon was designed to be flexible in the multiplier department, so knowledgeable overclockers could play around to their heart's content. This philosophy has been extended with the Thunderbird. Remarkers are thwarted because (A) the chip's rated speed is printed in plain sight and (B) modifications to what we hope are the multiplier control circuits will be easy to spot. At the same time, if the aforementioned press release is to be believed, overclockers will be able to manipulate the multiplier via motherboard jumpers.
All this keeps the overclockers happy but makes life hard on the remarkers. A vendor can simply glance at the top of a chip and tell if it has been monkeyed with; there's no risk of hack modifications hiding under a plastic casing. While it's true that an unscrupulous vendor might sell motherboard/CPU combos that hide modifications under a heatsink, such activities can be traced back to their point of origin with little trouble, which is AMD's main concern.
So assuming all of the above information and speculation pans out, I think the Thunderbird is going to be a fantastic chip for the overclocking crowd. AMD's happy because they're getting our bucks instead of Intel, and we're stoked because we don't have to screw with overclocked AGP busses and choosing between end of life cycle chipsets and memory translator hubs. Everybody wins. Except Intel.