Edit: on second thought, my initial answer wasn't very good. Here's another go with better sources.
For the Mars 2020 rover, one of the science goals is to collect 20 samples. The rover has room to collect 31 samples.
In the initial design, the rover was to carry all its samples on board throughout the mission. But then once the rover has collected 20 samples, the question becomes, what to do with them?
- If you keep going to collect more samples, there's a risk the rover will break down somewhere inaccessible, and you won't be able to retrieve the 20 samples.
- If you end the mission at 20 samples, you might miss out on valuable information.
So JPL has come up with a plan where the first 20 samples will be dropped off somewhere accessible, leaving the rover free to go on an extended mission without endangering the 20 samples.
The sample return mission is not being designed in detail yet. For now, the number of samples to be returned is set at 31. But it's possible for the rover to carry more than 31 sample tubes, giving the scientists more choice: if sample 32 turns out to be really interesting, they can ditch one of the older samples. Current thinking is that the rover will carry 40-50 sample tubes.
With the old-style cache bucket, you hit an interesting problem once you reach your minimum success goal of 20 samples. By definition: the mission is now a “success.” You have a treasure trove of samples that you could set on the ground for future collection, but you could also fill it with 11 more samples of indeterminate importance. You don’t know what or how important these samples could be, since you’d have to continue looking for interesting rocks to drill. But every meter you rove carries with it risk: risk of breaking down, risk of getting stuck in sand, risk of a computer malfunction, you name it. Despite JPL making it look easy to operate a rover on Mars, it’s not.
So there would be an enormous amount of pressure to drop the cache container once minimum success is reached. Do you cash out your winnings or let it ride? Even if you kept going, the levels of risk that the team would accept for roving to interesting places would be very low. This risk-aversion would increase with every sample collected beyond the minimum success point.
With the adaptive cache (or cache depot) strategy, you don’t have the same problem. You just tell the rover to drop the 20 samples on the surface, effectively creating a cache depot at that spot on Mars (the rover can carry a certain number of sample tubes with it while it drives). The rover then continues on to other interesting places where it can drill additional samples, safe in the knowledge that it has already deposited a successful mission’s worth for future retrieval.
There are additional practical benefits for the cache depot strategy. Since Mars 2020 would no longer be limited by the size of a cache bucket designed to return every sample to Earth, the rover could bring along more sample tubes to use throughout the mission. And since the samples are deposited frequently, which frees the rover to go on in search of more science, the team would have more time in which to debate which samples to return. Without this, the team would have to decide far more quickly about where to drill with less contextual information about the overall geology of the landing site.
This design isn't final yet, JPL are still working on the design concept. But several cache strategies have been studied, with adaptive caching turning out to be the preferable strategy.