128 bits accesses data 16 bytes at a time
256 bits accesses the data 32 bytes at a time.
so if you had to get 1kb of data the 256bit engine would only require 4 fetches instead of the 8 for the 128bit engine.
On top of this you need to factor in DDR type and speed.
As for recommendation, I'd go with a gtx460 myself. The 768mb version is about equivalent to the 5770 and the 1gb version is about equivalent to the gtx465.http://www.tomshardware.com/reviews/geforce-gtx-460-gf104-fermi,2684.html