There is quite an interesting thread on the Goodreads.com forums about this topic.
Goodreads have chosen not to do it, and state some of their reasons why in the thread. Basically it comes down to psychology of choices and that Netflix apparently tried it and found they got less reviews from users who thought that the number of extra options they now had to choose between (from 5 to 10) made them not be bothered to rate at all.
Personally I would be in the "like to have half stars" camp too. But in reality I really just want to have the option of a "3.5" and a "4.5" - I wouldn't ever use 0.5, 1.5 or 2.5.
Since 0 is unreviewed that means the minimum you can give a book is 1 star. I rarely do that because to me it implies the book has not one single redeeming feature - it is the "worst of the worst". It has happened, but rarely for me. So the normal "book was poor" rating I will give is 2 stars.
5 stars is a book I thought was absolutely outstanding. So that just leaves me 3 stars and 4 stars to fit every other book I read into. So 3 stars became "it was ok, a few things bugged me", and 4 stars is "really good, but not an all time favourite". The numbers themselves don't mean anything of course, but a 3.5 would allow me to differentiate the gap between them.
However as Calibre doesn't support it, and neither does Goodreads, I live without it and spend more time reading instead of fretting about it.
The most time I spend with ratings is "correcting" a rating after metadata download due to the rounding/averaging - that a book with a 3.98 rating ends up as 3 stars instead of 4 is something I feel compelled to "correct"...