speculating here ...
it is possible that it would represent the difference between sga_max_size and sga_target? that space would be reserved in anticipation that you might need it to expand? A quick test would be to increate cache_size by 128M and see if that makes a correcponding dent in KGH-NO ACCESS.
while on the subject of speculating ... doesn't kgh functions correspond to heap allocation/deallocation ??
Of course I could be wrong and would welcome corrections. Raj