The test document you send us to illustrate HELPD-249 hits performance limits of our CSS cascading engine (complex CSS constructs need to be applied to a big document). PD4ML v3 had a simplified support of CSS selectors, so quite a lot of CSS rules were silently omitted – that explains the performance degradation.
As you see, we’ve already done some CSS engine optimization (4.0.9fx4 vs. 4.0.9fx3). But a more serious CSS cascading logic revision on architectural level is planned to v4.1.x. We returned to work on v4.1.0 few weeks ago.