Once a PLL has "locked" on a signal, it will track the signal provided there are nondiscontinuous phase changes, and the signal remains within its tracking range. In the worst of cases, a PLL will take quite some time to lock on a signal even though the signal frequency is within its capture range depending on the phase of its internal VCO at the time in relation to the phase of the input signal (also on the characteristics of the LPF). The result of this delay is that the detected pulse width is smaller than the actual pulse that was sent, by a random (but bounded) amount. The bottom line is the data rate suffers. It is quite possible to send a tone burst, and detect it by using a tone detector such as the 567, but if the tone burst duration is small, sometimes (worst case about 14 cycles) the 567 will not detect it.
Anyway, this is at least my understanding of it. I haven't been a commications engineer since four jobs, and about 16 years, ago!
Jem