I wonder how many of the 60% non-unique keys are embedded devices with hardcoded keys...
That wouldn't surprise me much at all.
I've been working on a secure workaround to let an embedded webserver on our consumer-oriented wifi-equipped hardware(1) do OAuth2 authentication (to connect to owner's Twitter/Faceboot/et al. accounts). OAuth relies on SSL/TLS to protect it's password-equivalent tokens on-the-wire. Getting an SSL cert onto the device (at a reasonable cost) so that a phone connecting to it via wifi doesn't throw up big scarey "somebody might be doing _bad_ things!" warnings is not easily solved.
(1) Shameless self promotion: http://holiday.moorescloud.com - The world's most intelligent christmas lights…
Sigh. I'm not sure what we can do (as an industry). Bad devices (and bad firmware, and bad software) are so widely deployed it seems almost impossible to fix.
If we're talking 1024-bit integers created by multiplying two 512-bit primes, the facts are:
1) The number of 512-bit primes is estimated around 11.36×10^150
2) The chance that two truly random primes are the same is one in 3.37*10^75. It is safe to assume it has never happened and never will (at least on our current hardware).
I find the scale of these numbers absolutely incredible.