I don't have the link, but I'm with your friend on this one... Something smells fishy. If this were a legitimate, lossless compression scheme that worked on general data, it would revolutionize everything. If you find the website again, look for details about this data. I would guess it's a specialized compression format for specialized data. Here's a classic example:
Run-length encoding: you look at a file, count the number of bits in sequence that are the same (either a sequence of 1's or a sequence of 0's). Then, write how long that sequence goes for instead of the sequence itself. In other words, something like this:
Code:
original data: 1111 00000000 111 00000 1 0 11 0000
run-length encoding: 4 8 3 5 1 1 2 4
This compression scheme isn't so good for general data, but I could make a file that caters to the format. Imagine a set of files that total 4TB worth of 0's. With run-length encoding, all of those files could be compressed to a file 6 bytes in size. For me to claim I have a compression scheme that reduces 4TB of data into 6 bytes is misleading at best.
I'm not saying it's impossible, but the devil is in the details. Snoop around that site if you ever find it again.