Just to be clear, IMO rectification is the dominant process that suffers from speed issues (gain bandwidth) at low level not the logger.
I became all too aware of this designing the dB conversion for the TS-1 back in the early '80s. The vast majority of rectification circuits log or linear, require the active (opamp) gain stage to swing from - to + a full diode drop at every zero crossing as the signal passes through 0V dc. Generally opamps that start out with 100dB of open loop voltage gain, this input voltage to make +/- .5V at the output is no big deal. However at higher frequency as this open loop gain gets rolled off for stability compensation gain becomes an issue.
To put numbers to this, Say your general purpose opamp has been compensated down to 60db of open loop gain by 20kHz. This means at 20kHz to make the 1Vp-p output swing at zero crossings, the input needs 1mV of drive. This 1 mV is effectively subtracted from the input signal. For a several volt signal this 1 mV is not missed, but for an input audio signal that starts out with only a few mV, subtracting one mV is significant. In practice this looks like a LPF rolling off the audio but instead of just being frequency dependent, its affected by both level and frequency.
This is not a new or unknown problem so there are sundry circuit techniques to improve this metric. For me it was a matter of concern because I was making test equipment so little things like frequency response errors were undesirable. In my experience for dynamics processors the frequency response of the side chain at -60dB is not a big issue. It's not like it stops working at -60dB but the HF response falls off. I found when designing companding NR paths for tape recording or effects, it was often desirable to frequency shape the side chain to avoid those pesky extreme frequencies that don't always survive the media (tape, BBD delay, etc).
Getting back on topic, the response of a logging junction will be somewhat affected by current density, so that becomes a tradeoff to get useful dynamic range. Too much current causes errors at the high end due to Rbb resistance, which adds a linear term to the log output voltage. In my TS-1 the rbb linear error was almost 1dB at +20 dBu.
I suspect the RF chip is using an alternate approach. I would need to see more details about the inner circuitry to comment intelligently but I can tell from the block diagram it isn't a simple log conversion. So yes it promises to be faster, I am still unsure that this is useful for audio processing.
@ Mr. Coffee methinks your problem is more fundamental. Simply stated you can't detect a signal transient with good certainty until that transient has actually finished happening, or enough of it to be confident. One design approach to deal with this for a noise gate (RANE) was to put the audio path through a short delay, that way the un-delayed side chain could have a few mSec of look ahead, more than enough time to open a gate and not compromise the leading edge of the transient.
To make a guitar effect that ignores the initial pick, it seems a simple delay in opening the gate should do that. The speed of the side chain is not that much of an issue if you plan to delay the action.
JR