i noticed this too. as glamor mainly uses gpu to do all rendering i think the gpu / mali drivers are bottleneck here.
i checked with some odroid xu3/4 (similar mali gpu) developer memeka who helped me before with machybris : see http://forum.odroid.com/viewtopic.php?f=95&t=18905#p149215
he would expect 5-10 times higher scores for glmark2-es benchmark! so our glmark2-es scores should be 1000+! however, i dont have the actual results for an odroid xu3.
but if that is true then on rk3288 there is something inefficient. that might be related to ineffecient synchronization, bad timing or bad configurations in kernel parameters in resource/dts file.
so my first bet would be that gpu is bottleneck for glamor...
however if glamor itself is the bottleneck i guess they could optimize this by using zerocopy for "copying" window frames into textures instead of the usual expensive egl/gles calls. so there is no data copy then but only memory sharing.
actually androids surfaceflinger (display manager) is doing this.
and jacob chen has doing this for the vdpau drivers.
as for rga, actually i had got drm rga running before on firefly kernel 4.7 by porting the patches i found at rockchip repos.
cheers -- mac-l1