Performance become very slow when using multiple threads to do net_->Forward();

[TianBinyu]

When i am running caffe in android platform, the performance become very slow if i use multiple threads to do prediction, if there is any solution for that?( CPU only~ not GPU)

Thank you very much
ps:
if i use one thread to predict 1 image, it will cost 100 ms. 4 image will cost 400 ms(first prediction will cost 300-400 ms)
But when i created 4 thread to predict 4 image at same time, one image will cost more than 500-600 ms... :(