If you look at how keychains are usually used nowadays, they are an implementation detail you could and probably should hide from non-technical users.
As you say yourself, in order to find a key, you use the search for key button. Which prompts the program to look on a keyserver. You then import the first non-expired key you find for the desired email address. Why not do this automatically? You just annoy users who have to go through two (highly non-self-evident) steps and who do neither know nor care what a 'key' is. They just want to write secure email to their friends. (and btw, does the key-search even go over a secure connection with a pinned certificate? Because else you are not even trying to avoid MITM attacks).
Is it too much to ask these users to read the manual, learn about which kind of key is the most secure yet also compatible with the GPG versions of their friends, find out how many bits theirs should have, have them wiggle their mouse (why, couldn't you use /dev/urandom for most cases?), upload their key to a keyserver, search their friends key from a keyserver, learn about the web of trust and go to a key-signing party? Is this too much to ask? Probably yes. I did not need to acquire a similar amount of knowledge about TextSecure and yet it probably transmits my messages in a more secure form than GPG.
I am not saying that key-discovery is super-easy and all these are completely solved problems. But why do you have to make it so much more complicated for users than it has to be?