Thanks you two!

I scale now the movement vectors before using them and removed the time_step multiplication for PL_MAXGRAVITY and PL_JUMPACC. It seems to work now. But I still don't understand why I had to keep it for PL_GRAVITY.