Mali's slides here still show them doing two vertex shading passes, one for positions, and again for other attributes. I'm guessing "memory" here means high-performance in-unit memory like TMEM, rather than a full frame's worth of data, but I'm not sure!