I have dual gtx460's watercooled & oclocked to 851mhz and they hash at about 145 Mhash/sec (70Mhash/sec each)
This is the norm for nvidia.
 

I use two rpcminer-cuda.exe instances like so:
-gpu=0 -gpugrid=320 -gputhreads=320 -aggression=7 -workrefreshms=1000
-gpu=1 -gpugrid=320 -gputhreads=320 -aggression=7 -workrefreshms=1000

Guiminer Messages about "lost connection" are normal, its from the pool servers reply time.  

Every server has different reply times, if your miner tries to query the server too soon, guiminer will say you have connection problems.
You need to find the refresh rate your pool allows and change -workrefreshms=XXXX to the servers limit.
Default for  -workrefreshms=4000


Most virus scanners and firewall programs hate "coin" miners and call them viruses, so make sure your virus scanner and firewall allow rpcminer-cuda.exe, and guiminer.

I use p2pool as my pool server.
https://bitcointalk.org/index.php?topic=18313.0

DO NOT USE rpcminer-mod-cuda.exe
This MOD also uses up all of 1 cpu, so i think its using the cpu to do hash also, perhaps tricking users into thinking its better than original.  It uses all cpu, and no hashes are accepted in GUIMINER, even tho it says its hashing, and p2pool says almost all are DOA...
DO NOT USE THIS MOD-->hxxps://github.com/cdmackie/rpcminer-mod/tree/master/bin
 USE ORIGINAL RPCMINER

Original RPCMINER: https://bitcointalk.org/index.php?topic=2444.0

If your RPCMINER program uses 100% of any cpu, and GUIminer shows 0(0) under "Accepted"
 (if you are only doing gpu mining) then you have the modified broken rpcminer-cuda.