HP magic

I'm trying to understand something. One of those things HP did on their equipment long ago...
Referring to this: http://www.hpl.hp.com/hpjournal/pdfs/IssuePDFs/1976-12.pdf diagram on page 10 explanation on page 11.
More info here: **broken link removed**
There's even a schematic for the special chip somewhere on the net but I don't have a link to it.
I've been simulating this comparator (LM339) and peak detector (BAT54 + 12pF) combo in LTSpice, and assuming they've shown it connected how it really is, there's no useful improvement - I can get a slightly extended response with an oscillating input (improved to 12.5MHz from 11MHz) but it does nothing to help single pulses, which get lost in the cap.
Any suggestions? Should the improvement really be so slight?

I assume that you are doing something like this:


My first thoughts are:

The 47pF and the 12pF capacitors are making a capacitive voltage divider.
This will affect the voltage sensitivity of the comparator system.

The 130 Ohm resistor and the 12pF capacitor are forming a low pass filter.
This low pass filter action will limit the circuits response to fast pulses.
Try reducing the value of the 12pF capacitor.

JimB

PS
Many years ago at a place where I worked, there was one of these HP IC Troubleshooting Kits as shown on page 17 of your link.
I thought that it was a very desirable object in its nice soft case.
I wished that I had one.
Fast forward many years and I came across one for sale at an affordable price.
As I was debating with myself whether to splash the cash, it occurred to me that in those intervening years I had managed very well without one.
So, money back in the pocket and no regrets since.

HAH!
What they don't tell you is, use different sized capacitors! That is the magic part...
I'd arrived at 12pF as a compromise after settling for 15p for a while then dropping it down a bit. Actually the "high" comparator needs a smaller cap, but the "low" one needs a bigger cap. So 4.7pF and 18pF on it now.
[further experimentation]
The high comparator will perform up to 50MHz if I drop the cap to 3.6pF. The low side one stops being useful past 20pF and will work up to about 37MHz.
This is all based on pulses with 50% duty and 1nS rise/fall times
Change it 0.5nS and I have to change the caps to be closer in value.
To be honest I doubt if the rest of my circuit could keep up anyway - it's all based on a single LM339.

I don't understand why the peak detector improves the response of the comparator (to a clock signal) though. Anyone?

OK further update on this for anyone who's interested. I finally got (in simulation) a LM339 (1.3uS) comparator to trigger with a 10nS pulse! (actually got it down to 1nS but I was only aiming to match the 545A's performance). The choice of Schottky diode makes a huge difference - even between different RF types. So the one that's plugged into LTSpice at the moment is the HSMS-2820. It all comes down to reverse-leakage. The caps still need to be different sizes though and the logic-0 detector side needs to have a pull resistor switched in or out for the CMOS/TTL settings to both work, which is annoying.
But, yay!