Hi shMatrix, to optimize even further, in the next build I've disabled some kernel options used by developers (aka debugfs).
This can improve something, but I don't expect miracles.
My suggestion is to try with another image considering that the one for mxq4kpro is for rk3228a. This soc in teory should have a lower clock for cpu and gpu (see first post for details).
Often these boxes are sold as a model, but internally they may be different.
For example mine is an mxq4k, but the board is identical to the one present in v88 mars
So check in the first post if there is an image that is similar to the one you use(ddr3, same wifi chip, same internal memory etc etc), but suitable for rk3229 / rk3228b and download it from my gdrive.