The amount of data that an individual FDR contains, while dependent on the particular unit, is fixed, and would correspond roughly to some amount of flight time. How far into the past that extends, and how many flights it captures, depends on how often the aircraft flies and how long the flights are. If aircraft 1 is operated for a single long flight every few days, and aircraft 2 makes 10 short flights daily, the memory in the FDR in the second aircraft would contain more flights, but going back fewer days on the calendar, than a comparable FDR in the first aircraft (before the data is getting over-written).
At major airlines, and probably many smaller operators, the FDR is downloaded regularly by maintenance as part of a FOQA (Flight Operations Quality Assurance) program, and the data is stored so it can be analyzed for trends over time. This way, if some worrisome event comes to light, it's possible to have a server go back through data across a fleet of perhaps hundreds of aircraft and years of operations to see how often something similar happened, or almost happened but wasn't noticed.
Thus, to take the example from the OP, if somebody noticed that flight control checks weren't being performed correctly, a FOQA analyst could write a query to determine if a particular flight control check was "good" or "bad", and then have the query run against the data from tens or even hundreds of thousands of flights. If that found that most flight control checks were done correctly, they might go further to see if the bad flight control checks correlated with something... short turn arounds, or late night flights (i.e. tired crews getting sloppy), or something else.
Then, instead of a message to the crews "hey, make sure you do your flight control checks", the message could be more specific: "Even on quick turns, it's important to make sure you get full travel on the controls, because we've seen XXX occur even when the aircraft was only parked for a few minutes..." That sort of a specific, focused message tends to affect behavior more than "Hey-- do your job right, darn it!" messages do.
Thus, data like what the NBAA referred to in the OP doesn't necessarily have to all be contained in the FDR; it may have been downloaded & stored over time instead. Data for 175 flights from one aircraft seems like it could have been all in memory; if that were a couple hundred hours of recording, that doesn't seem like an unreasonable amount of memory to put in a FDR. On the other hand, the analysis of data from over 100,000 flights over two years is almost certainly from data that had been downloaded regularly over that span of time.