looks like you're probably trying to crack two handshakes at once instead of one. remember that v1.00 added multi-cap support so you need to clean your cap file to ensure you're only cracking the one that you need.
When I crack my .hccap file, it show:
Recovered.....: 0/1 Digests, 0/1 Salts
if I had more than 1 handshake, it doesnt show 0/1 Salts but more (e.g. 0/10). Or not?
this does not make any sense, you should definitely show much better performance with -u 4096 than -u 1024. the explanation of these parameters has been covered dozens of times before, but to re-iterate, -u is the number of kernel loop iterations. wpa uses 4096 rounds of pbkdf2-hmac-sha1 so 4096 will naturally show the best performance since the loop iterations match the number of rounds. except not with your card, apparently, and i'm not sure why. it might be because it's such a low-end card.
It might not be hashcat but your Nvidia card drivers that are making the difference. I've noticed a double in wpa cracking speed when using Nvidia driver versions between 320.00 and 327.23 any drivers before or after drop my speeds down from 100k to 50k. Why the newest drivers are slower I haven't got a clue. Also see http://hashcat.net/forum/thread-2269.html