Okay, I'll chime in...
The purpose of reinforcement is to increase the occurrence of a behavior.
One of the main principles of ANY learning theory (whether it be classical or operant conditioning) is that they, for the most part, follow Aristotle's Laws of Association, of which there are four. Now, many behaviorists maintain that the laws of similarity and contrast don't play as major a role in conditioning as do the other two: the Law of Contiguity and Law of Frequency.
Law of frequency is easy to understand: the higher the frequency in which two things are linked together, the more likely they are to be strongly associated.
Law of contiguity predicts that things that occur close together in space and time, will be more likely to be associated.
Now,take these laws and apply them to learning theory. The more frequently you pair food with a behavior, the more likely the organism will be to perform the behavior. Why? Because the behavior has, in the past, brought about an appetitive consequence.
But, you have to also take into consideration the law of contiguity. Things that occur together in space and time.
So, if the dog gets the treat, right after it comes to you, then you are positively reinforcing THAT behavior. In reality, this is how many of us teach a recall to our dogs. We call dog, dog comes to us, dog gets treat! So, why would we even say that we are "resetting" the dog by getting it to come to us, when, in reality, you're just teaching it to come to you.
Now, is there such a thing as delayed reinforcement? Yes, but there is not a ton of empirical support for the idea that it is as effective. In fact, most researchers cite Skinner's (1938) original study using immediate reinforcement and reinforcement occurring at 2, 3, and 4 second delays in their arguments AGAINST the efficacy of delayed reinforcement. In fact, Skinner found that while the rats COULD acquire the behavior when reinforcement was delayed (2, 3, and 4 seconds), it was severely retarded as compared to immediate reinforcement.
Now, if we look at marker training (which is actually referred to as a conditioned reinforcer by behaviorists), many studies have shown that the conditioned reinforcer functions in much the same was as a primary reinforcer, where a primary reinforcer is simply a reinforcer that is inherently good, and the organism does not have to "learn" it's value (the most common examples of primary reinforcers are food, water, and sex). A conditioned, or secondary, reinforcer on the other hand is one in which the "goodness" of it is learned (for example, the clicker in clicker training) through classical conditioning (the pairing of a previously neutral stimulus with an unconditioned stimulus)...However, this only occurs if the proper classical conditioning trials take place for that previously neutral stimulus.
As a side note, seconary reinforcers are also the primary mechanism in our economic system (i.e., you aren't born knowing that money is good, but you do learn that it can be used to get you other things, which is good) which is basically a larger scale token economy (commonly used in mental hospitals and prisons).
Therefore, to me, you don't need to use this "reset" with a dog if you're also using other consequences in your training with the right frequency and contiguity...and, no, people, I'm not simply talking about a prong collar correction
...even something as simple as turning your back - negative punishment - used in conjunction with positive reinforcement can gain the behavior you want (I've done all of Axo's training using these two consequences, and Greg has even made a comment that he's starting to look incredible with his behaviors, all things considered (his age, etc))...
See, this is what happens when you catch me while I'm writing a lecture for learning!