First of all, I was comparing a PFD to a _good_ phase detector (DBM or MUX) and not to an XOR. However, XORs are in fact generally quieter than PFDS.
PFDs have a lot more going on than XORs, which means more jitter and drift, and (unless you're doing something fancy) the servo point is in the middle of a deadband, so it hunts around like a bastard within about a 3-ns window for HCMOS or wider for metal-gate CMOS. The deadband is caused by the output becoming a runt pulse when the timing error is less than the sum of the rise and fall times of the output. The pulse amplitude becomes sensitive to supply variations, small temperature changes, what you had for breakfast, and so on.
The loop filter won't let an XOR lock up on the wrong edge, because the DC feedback is positive there.
Pure nonsense. To even begin to use a PFD with a noisy signal, you have to use a bandpass limiter such as an FM IF strip just to make logic levels. That degrades your SNR by intermodulating all the noise components, which may make locking impossible all by itself. It's also useless for signals where you don't know the centre frequency that accurately.
Also, with a PFD, if you miss a transition due to noise, the loop unlocks and has to re-acquire, whereas an analogue PD or even an XOR just carries on. That alone makes PFDs completely unsuitable for noisy signals. A narrow loop filter is no help whatever for either of these problems--by the time it sees the signal, the damage has already been done.
Cheers
Phil Hobbs