Joint Stereo doesn't blur the stereo image. It's just a lossless transcoding from Left / Right Stereo to Mid / Side Stereo. Don't mix it up with MP2 Joint Stereo!
As in most musical pieces the content of the left and the right channel is very identical this saves a lot of encoding bandwith and gives you a higher quality at a given bitrate. As you can mix M/S and L/R stereo frames in MP3 files, at least with the LAME encoder "Joint Stereo" means that the encoder automaticaly decides if it is better to use Mid/Side coding or Left/Right coding for a MP3 frame to save bandwith and give you the best possible audioquality. Forcing Mid/Side or Left/Right (which is both possible) is in most cases braindead and should only done when some very old and buggy decoders are used.
ABR is somewhere in between CBR and VBR. It uses the CBR psychoacustics (which aren't as good as the VBR ones), but tries to match the target bitrate while being flexible with the used frame bitrate. Silent parts (at the beginning and end of the track) can easily be encoded with a very low bitrate (like 32 kbps) but you will certainly have parts where a fixed bitrate (like 128 kbps) isn't enough to encode without artefacts. So the ABR saves bits where it is possible and uses the saved bit it places where it is needed.
All three models have their use:
CBR: For streaming where you need a fixed bitrate
ABR: When filesize matters, but bandwidth changes aren't problematic
VBR: For all other cases (Best Quality at least possible bandwidth)
IMHO the OPs question can't be clearly answered. It depends on the music and your listening equipment. If have lots of music that is transparent at 128 kbps, some that is transparent at 160 kbps, some that is transparent at 192 kbps, some at V3, some at V2, some at V0, some at V0 -Y, and some that needs to be encoded lossless because I hear artefacts no matter what encoder settings I use.
I normaly use V0, and the lossless codec of my choice (WavPack) for stuff that still isn't transparent. But I must mention I listen to a lot obscure stuff like Industrial, Noise, Power Electronics, Dark Ambient and Neofolk which is hard to encode lossy. For people with normal listening habbits V2 or V3 should be enough.