I am attempting to debug a peak detector circuit. I'm feeding it with 20 us pulses at 480 Hz. Above a certain voltage, the circuit starts "missing" pulses. Rather than charging the output capacitor every cycle, it charges it every few cycles. This increases the ripple voltage rather dramatically.
This is the basic circuit I'm using. In my case, R1 is shorted, R2 is 1k (I'm not entirely sure what affect this has), diodes are 1N4148s, U2 is a TL071, and I've tried everything from a TL071 to a LT1226 for U1. The circuit is powered from +/- 12V.

Here's what one of the pulses looks like up close. Yellow trace is input, blue trace is output. The difference in vertical scale is because the input is terminated into 50 ohms (thus cutting the amplitude in half), and an output stage downstream of the above schematic has a gain of 2. The signals on the scope screen are scaled to compensate.

Here's what the output is supposed to look like, and what is does look like when the output voltage is below around 2V (the max voltage for good behavior seems to vary a bit).

Here's what the output looks like when there's a problem.

Any insight as to what's causing this would be much appreciated! I usually do mostly digital work, but I figured "how hard can a simple peak detector be....". [sigh]