Correct, the external clock in a dCS system does not replace the internal clock in a dCS unit, such as a Rossini DAC; the external clock instead acts as a reference for the DACs on-board clock to work to.
With that in mind, you want to isolate your reference clock from any kind of possible interference. This includes electronic noise, power rail variations, temperature variations and physical vibrations. All of these factors can influence a clock’s performance.
Think about everything a DAC, Upsampler or Transport has to do. They’re pretty big jobs all things considered, with lots going on simultaneously inside each unit. This will generate heat (from things like the amps), small variations on the power rails, and potentially very small amounts of crosstalk between components.
While we do an awful lot of high-level engineering to minimise the effects of these factors in these units, an external clock naturally has much less to do. It has a power supply that is not feeding amps, transports, processors etc., so the power fed to the crystals in a standalone clock is cleaner, they have more stable temperatures, less physical vibrations and no DAC boards, amps etc. inside the same box.
This isn’t to say the on-board clocks in a Rossini DAC for example aren’t already very good, they still have to be (as ultimately they are the clocks controlling the audio). But, feeding them a stable reference clock from a unit which does not need to worry about the external factors above will be, well, better! It’s essentially like removing all of these external factors from the clock inside every DAC, Upsampler and Transport connected to the external clock.
In a perfect world the clock inside let’s say a Rossini would live in a little vacuum with perfect power, heat and physical stability. In that case, an external reference clock wouldn’t make sense. Unfortunately though, you cannot cheat physics.
In practical listening I would wager most people who have experienced a dCS system with an external clock would agree this gives a very high level of performance. Subjecting the clock to external factors for the sake of not having a PLL to our ears is not the way to go! Besides this, the PLL (which as Anupchas touched on) is a very well known, understood and optimised bit of tech. We also do some clever stuff with ours to minimise jitter on the incoming master clock signal.