"Implement the standard" only works to the extent that there is a written specification of the behaviour. In practice, CSS specs already have a problem with insufficient reporting of bugs in the specs, perhaps in part due to insufficient resources for fixing reported bugs. One fewer rendering engine means both fewer bug reports, and (I suppose) fewer person-hours available for fixing bugs.
As for "What's wrong with having only one implementation", the problem is that not all uses of HTML/CSS are for web browsing: you want your word processor to be able to read & write HTML/CSS, and publishers want to use HTML/CSS for writing books, but there are fundamental conflicts between the needs of an interactive web browser and the needs of printing. (One needs to be fast and support interactive technologies, the other needs to look good: well-chosen page breaks, line breaks, float placements, table column widths etc.)