Another aspect is that with 3k samples/sec you will sample the AC in 30 'slices' ~ 3% of the width so the max measurement will be 0-3% off the peak amplitude. and 3% of 256 being 7.7 will give you another 2-3 bits of error.

True, with 3KSPS I will be getting 30 samples per AC cycle i.e. at each 3% cycle interval. I am aiming at getting each sample as precise as possible 7 bit is really a dream though. When the RMS current is calculated from these sample certainly the RMS calculated will be hardly any precise. But In this application upto 5% error in measurement of RMS current is not a problem. For resistance welding 5% difference in current doesn't make any significant diffrence in weld quality. Still I want to try my best so that I can keep it as accurate as possible.

Regards,
Prahlad Purohit