However, here are a few pointers for people who are new to CSS, struggling to get CSS positioning:
http://www.alistapart.com/authors/s/nstokes
Another very important issue is that of "containing floats", http://complexspiral.com/publications/containing-floats/ , th that the height of a container is not expanded to fill its contents is logical from a publisher pov.
Where does one start the next paragraph if a paragraph has a long image on the right? It doesn't start when the image ends, the next paragraph starts when the text in the previous paragraph ends.
Still, though, CSS positioning is hard, and that's why people are using grid frameworks, like 960.gs.
One other major issue is CSS is that all browsers have an internal stylesheet, and those sheets are slightly different, hence people make use of CSS resets or normalizers that default properties to common values.
As a sidenote: To see how one can use CSS to build quality documents, look at PrinceXML: http://www.princexml.com/samples/
Pixel perfect on a web browser doesn't make a lot of sense since people could resize their browser, increase/decrease font size, etc.
That being said, CSS has always been kind of a black art to me. Floating, em vs percents vs pixels, rules regarding inline vs block elements, the box model with borders, padding and margin... These are all things I had to read up on and experiment with to figure out how they worked. I've spent quite a while trying to read up on all of these things to figure it all out. Contrast this with learning layouts in Java Swing, QT, GTK+ in about 1 days time each. I realize I am comparing a GUI toolkit with the web but... something seems broken here. I'm not sure if it is lack of clear documentation and tutorials, inconstant implementations in older browsers or a failing of the language itself. Maybe its a combination of all of those.