Well, besides the fact that you would need to block TCP as well as UDP (RFC calls for support on both and longer messages, such as zone transfers, require TCP due to UDP's content length limits), you also have the benefit of the fact that this would block exploits that make use of port 53 for communication on the strong likelihood that it is completely unfiltered.
The AS article asks where is the best place to filter though. This gets tricky. The request doesn't indicate whether this is enterprise equipment or consumer. The mention of router-based filtering implies consumer though, so I'll focus on that.
First off, a good number of consumer routers do not have the processing power to handle full filtering at high speed. Even enterprise appliances such as iPrism require heftier units when the pipeline speed exceeds a certain threshold. As a good example, a Linksys 625 Wireless Router can handle filtering with no rulesets up to about 50-60Mb/s. Rules are relatively efficient, but there is no way in many cases to automate rule implementation, and when the ruleset increases in size the capability of the router to handle it drops to around 20-30 Mb/s. Fine if the WAN uplink is perhaps a 15Mb/s line, but catastrophic if you're trying to get full use from a 105Mb/s cable or fiber line.
The end answer really comes down to a balance. Implement filtering at the furthest end that you carry absolute sovereignty over, balanced by duplication of effort and complexity of implementation. Replicating rules over thousands of endpoints is complex enough and lacks enough control that performing the filtering at the trunk is more efficient and effective. By comparison, the ability to control one or a few computers in a home is substantially more likely and will take the burden off the limited processing power of a consumer router. Walking a rule manually to five endpoints is trivial compared to dozens, hundreds, or thousands.
If the uplink is small enough to allow filtering at the router and the eventual change and replacement of equipment will allow easy transfer of rulesets and administration, work from the router or a similar trunk location to globalize and centralize effort. If the endpoints are spread enough or there is sufficient lack of control over them to warrant such, again, work from the trunk. If enough trust exists in the endpoints to offload the work onto their substantially-stronger processors, and administration of rules to and of the endpoints is trivial, filter at the endpoints.