The accuracy and reliability of the canine sniffs are lower than many think. The privacy interest in the place to be searched is more important than the goal of those searching. I think dog sniffs of a residence are just as much searches as are thermal imaging devices which the Supreme Court in Kyllo held require a warrant under the Fourth Amendment.
Here's a good law review article on the topic, by Leslie Lunney. An excerpt:
The Fourth Amendment, and the personal rights which it secures, have a long history. At the very core stands the right of a man to retreat into his own home and there be free from unreasonable governmental intrusion.
Police employ drug-detection dogs in public locations, such as airports, as a quick means of determining whether luggage contains contraband. In United States v. Place, the U.S. Supreme Court explained that the use of drug-detection dogs to sniff luggage in a public location was not a "search" under the Fourth Amendment because of the accuracy and limited intrusiveness of the canine sniff technique. The Place Court likely reached this conclusion because the background understanding of the day was that detection dogs were the ideal sensing tool because, in the rare case of a mistake, the dog's error was actually a false negative. Therefore, any mistake by a drug-detection dog worked to the benefit of the luggage owner.
Despite recent evidence that drug-detection dogs are inaccurate a surprising percentage of the time, the Court in Illinois v. Caballes extended the warrantless use of the canine sniff technique to a lawfully stopped vehicle. The impact of the Caballes decision has been felt far beyond vehicle sniffs, however. Lower courts have taken the Place and Caballes decisions as a signal that canine sniffs are per se nonsearches and that it is therefore permissible to conduct suspicionless canine sniffs of homes.
A Chicago Tribune study last year detailed the accuracy problems with canine sniffs during traffic stops.
A Tribune analysis of three years of data for suburban departments found that only 44 percent of those alerts by the dogs led to the discovery of drugs or paraphernalia.
Another good read: Radley Balko's, The Mind of a Police Dog.
One more quote from Leslie Lunney's law review article, on the guess-work involved in the use of drug sniffing dogs:
Studies show that drug-detection dogs alert not to the illegal drug itself, but instead to a contaminant or by-product in the drug. In fact, detection dogs may not be able to detect the so-called ultrapure forms of drugs, such as cocaine and heroin, because of the extremely low vapor pressure of the unadulterated drug.
As a result,
With cocaine, for example, it appears that detection dogs do not actually alert to the cocaine itself because the drug is a topical anesthetic that “deadens olfactory senses.” Instead, the detection dog likely alerts to methyl benzoate, a high vapor pressure by-product of cocaine that can occur naturally or as a result of processing. Significantly, although methyl benzoate is a cocaine by-product, the molecule is also commonly found in everyday consumer products likely to be stored in a home, such as “solvents, insecticides, [and] perfumes.”
In other words, drug-detection dogs likely alert to an entirely legal substance, methyl benzoate, which allows the human police officer to infer that contraband is also present.