Email works fine, the problem is that lots of software implements it incorrectly.
Take spam filters for example: The correct way for this (and what I also do on my own private mailserver) is classifying mail during the SMTP session. Thus, if my mailserver considers a particular message as spam the sender receives an error message stating that his message could not be delivered.
And now take Gmail: Happily accepts most messages, but labels some of them as Spam. Of course, no-one ever checks the spam folder. A few misclassified mails and people will start complaining that email is broken.
Email clients have to do this because they use Bayesian filtering to get rid of Spam.
It turns out, if the sender of spam messages can find out which messages got through and which didn't, then they could build up the same bayesian filter on their computer, and figure out exactly what email they can send to not get marked as spam. Bayesian filtering only works if the spammer doesn't know which messages get through and which don't. I believe this is also one of the reasons images are disabled by default in gmail.
I'm not sure how this would work if a spammer targeted their own gmail inbox to discover which messages succeeded, but maybe spam filtering in gmail is only partially collective, so that individual inboxes build up slightly different bayes nets.
Take spam filters for example: The correct way for this (and what I also do on my own private mailserver) is classifying mail during the SMTP session. Thus, if my mailserver considers a particular message as spam the sender receives an error message stating that his message could not be delivered.
And now take Gmail: Happily accepts most messages, but labels some of them as Spam. Of course, no-one ever checks the spam folder. A few misclassified mails and people will start complaining that email is broken.