(Granted that in the most common use case, if I know the email address, I probably know what's in the inbox, since I am likely the person who sent the message. But still, why not make these 2 different random strings?)
A couple more examples: 1QjYwHNM vs 1QjYwHOc 1Qk07A9x vs 1Qk07A9X
You can verify this by repeatedly clicking 'Delete this address', which issues you a new mail box. The address string can be thought of as a number that is always incrementing, with the 'digits' drawn from [0-9a-zA-Z]. (So base 62.)
The URL number is always (sample size ~100) just a little smaller than the email number, but not by a consistent amount. Often about 20-30 steps away, but sometimes as much as 600 steps away (depending on server load?).
The number seems to be generated by a clock rather than a counter, since the rate at which the counter increases is very steady. The 5th digit from the right increments every 15 seconds, suggesting that there are (62^4)/15 ~= 0.99 million unique strings per second. Maybe the counter is just using the time in microseconds, and converting it to a string?
Issues raised:
1 - If I know your address, I can now guess the 'secret' URL to read your mail in a trivial number of guesses (certainly less than 1000). There doesn't seem to be any throttling of attempts on server side (I was able to manually cycle through ~30 invalid URLs fairly quickly).
2 - Email addresses 'leak' the time that they were created. (Not a clear security risk, but could be problematic in some cases)
3 - As Sami_Lehtinen says, addresses are guessable based on signup time, but given that these are disposable addresses, I'm not sure that's a big problem (who cares if a spammer guesses it). Also, even if there is a new user signing up every second, the success rate for guesses will be 1E-6.
