I remember someone mentioned trying a 3 dB attenuator in front, with no effect on oscillation. The oscillation is in the prescaler itself, and it doesn't matter how much gain is ahead of it.
The usual way to inhibit oscillation is to throw the input DC bias out of wack a little bit, with some resistance to Vcc, which is +5V for PECL operation, or ground for regular (N)ECL. This shifts the amplifier chain away from the "ideal" (Vbb) DC bias level, reducing the gain. It can be done experimentally, determined by the particular device input characteristics. A starting R value in the 1k to 10k range should be noticeable. You can monitor the input DC level with a high resistance like 100k tacked to the node, connected to a DVM. The loss in sensitivity can be made up for by more gain in the other amplifiers, if needed.
One thing that might help somewhat, is to use a much larger coupling cap into the input of the prescaler, so the Z there will be dominated by the driving amplifier's characteristics. It could also aggravate it, or just change the frequency - it's another experimental kind of deal. Bigger signal coupling caps throughout can also reduce the lower frequency cutoff, if that's desirable. Most prescalers can count as low as you want if the signal and cap are big enough, and edges fast enough.
Ed.
