Ontologies represent agreement. It become highly practical to develop useful ontologies
(a) depending the scope of the agreement (in increasingly more focused and targeted sphere, and hence more managable and feasible way: the whole world, common sense, the Web, discourse in a language, arbitrary topic that could occur in news, specific domain, specific task/application), and
(b) whether the agreement is capture as forma, semi-formal, and informal language/model (a viewpoint Tom Gruber drew my attention to).

Several highly succesful domain ontologies are already in extensive use, and for enterprise customers, more targeted populated ontologies are routinly developed (in our work at Semangix, to date, that has ranged from a million to 14 million instances). Those who still hold the view that it is not practicle to develop and maintain ontologies are either out of touch with avilable technologies and success stories in creating/applying ontologies or are tying to solve too big of a problem or a wrong problem for an ontology-driven solution. Ontology is not a pancea but for the right problem, it does wonders.

I have offered more extensive rebuttal to this point of view at:
http://lsdis.cs.uga.edu/~amit/blog/
Semantic Web: A different perspective on what works and what doesn't

Amit Sheth