5711 is a schottky, so the undershoot is capacitance. I like the SMS7621 because it's only about 0.25 pF. The tau against 1.3K is only a few hundred picoseconds.
We did, and it works great in several products.
The ECL bangbang PLL, the one we did for NIF, also works great. It has a ps or so of RMS jitter and temperature drift below 1 ps/degC. It was designed with Bode plots, never simulated.
We rolled our own no-deadband PFD in an FPGA. Rob did the PFD, in VHDL, and I did the diode-amplifier thing I posted, pencil on vellum. I think it was my suggestion to go with hard outputs, instead of a charge pump, when Rob started describing the risks of using FPGA tri-states to make a charge pump. I never liked charge pumps anyhow.
The other zero-deadband loop that would be easy is to make a hybrid, a circuit that's a frequency error detector out of lock and an XOR close in, sort of like the AD9901. Their logic diagram is on their data sheet. It looks like hell when it's out of lock, but it works. The loop filter could be a single RC into the VCO.
FPGAs can use counters and stuff to do the coarse lock, too.