I have spent last month trying to improve Painter, especially MT performance.
New tricks is increasing the number of paths that can be processed together, improving text rendering using new MakeValueTL thread local caching and unlike before, span fills (those with nonuniform color like images or gradients) can now be rendered at the same time (multiple span filled paths can be renderd concurrently with solid fills, before only solid fills could be mixed together). Also inition Clear command is postoned to this rendering phase, which should improve cache locality. As a result, some of PainterExamples are significantly faster in MT, typical is
In addition I have fixed a problem with Image placement being 0.5 pixel off (took 2 weeks to figure that out...) and add new attribute to specify advanced filtering kernels like bicubic or lancsoz 3.