A negative control for RNAi studies should always be another dsRNA, partly because dsRNA can illicit, amongst other things, anti-viral responses which you want to control for. If you were to use GFP dsRNA, you would have to be careful that it does not have off-targets to other transcripts (as you should check for all dsRNA's whether they are control or targeted). Generally speaking, GFP or LacZ dsRNA is a good control. Something else to consider would be the antisense of your target dsRNA.

When you say that the efficiency of knockdown is more with dsRNA control, do you mean that the level of expression is lower relative to the dsRNA control as opposed to the buffer control? Although it might seems as though the buffer is specifically reducing target gene expression, it is more likely that you are just seeing natural variation. Make sure you have sufficient biological and technical repeats.

A green fluorescent protein (GFP)-derived double-stranded RNA (dsRNA-GFP) is currently commonly used as control in RNAi or siRNA experiments. But, recently few authors reported off-target effects of GFP in aRNAi experiments (see link). This complicates the things and throws a few challenges to work on before experimental setup and analysis. 

http://www.mdpi.com/2075-4450/4/1/90