Projects like this would happen. There is a little talked about development called the BlueGene/P which is suppose to dwarf IBM's BlueGene/L.
The BlueGene/L will run at around 360 TeraFlops. (expected within a few months)
The BlueGene/P will run at around 1 PetaFlop. (expected within 2-3 years)


It is likely that before this particular protein folding project is completed both of the new machines will be operating at top speed.

I'm not sure how long it would take a PetaFlop supercomputer to calculate this little project, but I'd be willing to bet it would complete the entire project in the time it takes my 2ghz box to complete just ONE unit.



It is my personal assumption that the ratio is closer to 5:1.
Please keep in mind that in order for this grid to be effective and accurate it needs for each 1 unit to be processed on an undisclosed number of machines. It is this duplicate processing and cross checking that is a main contributer to the difference in overhead.
Obviously a supercomputer would have it's own error checking functionality, but it does not need to be as redundant and slow as the grid technology.
Simply put...If a supercomputer needs to check it twice (while already in memory) and the grid needs to check it 5 times (while in different physical locations) it stands to reason that the ratio of performance results is at a bare minimum of 5:1.

It may very well indeed be as close as 2:1, but like I said..I wouldn't bet on it.