Just wanna bump this up... I did a few ABX tests yesterday using the mac ABXTester app. I used either FLAC or ALAC for the lossless files and 320kbps MP3s created using Max for the lossy. I started with the opening track from the new Sun Kil Moon album, it has a plucked guitar line. My sense was that the lossless version had more space around it and a more 3D sound. I ran three rounds of of 5 ABX tests and got 60%; 100% and 60%. I then tried the opening to positive tension by bloc party. I ran that once and got 100%.
I was quite shocked by that, I was fully expecting to not be able to pick the difference. I realise that is a very limited sample and there may have been problems with the testing. But for me that result is enough to convince me to keep my music lossless. I reckon that there is a noticeable difference, which is subtle. But when so much of our gear is about accentuating the space and 3D imaging of our music, it seems sensible to me to stay lossless.
For statistical purposes it is better to run at least 10 trials in each set , 3/5 is not statistically significant and even 5/5 could be lucky guessing whereas 9/10 is pretty solid !