Selection cannot act over a single generation, if by 'act' you mean that an allele fixes in the population (reaches 100% frequency). If a new mutation appears in a single individual, it cannot in the next generation be present in every individual, unless the founder's progeny replace everyone else's. In any event this would be a disaster for overall fitness, since the level of genetic diversity would collapse.

In general selection propagates a variant exponentially fast based on the reproductive advantage it confers. If you make 1000 progeny and I make 1001, my reproductive advantage is 1.001 - every generation the frequency of my variant relative to yours goes up by this factor. If I start at frequency f and my advantage is s, it will take n=ln(1/f)/ln(s) generations to fix (approximately). For the example above with f=1%, this gives about 4607 generations.

There are other factors, notably population size. For a population of size N, the initial f will be 1/N. If N is small, f will be larger and selection will take less time. On the other hand, in a small population, genetic drift is stronger, and weakly selected variants are more likely to be eliminated by drift before they can fix.

I think it's important to note that evolution is really just the change of the availability and distribution of heritable traits in a population over time. To me, it seems fairly obvious that there are slow and fast components that can effect this, and therefore evolution plays out at multiple timescales simultaneously.

Your example is one of pretty fast restriction of traits: The trait of not being able to swim pretty quickly disappeared. However, how long did it take for the ability to swim to be present in the population in the first place? How long will it take for individuals to exist without the ability to swim after the event?

Yes, makes me think that maybe evolution can be looked at as operating in two ways, temporal random variations that aren't a guaranteed to be long term directions (presumably the genes passed in this test could just as easily breed out) and the fossil based evolution that we understand pretty well.

Of course IANAEB so maybe someone can give us a more detailed explanation...