As Pingu wrote, this is indeed interesting! Couldn't actually believe my own eyes. Still don't..
I've become very cautious of trusting the computer screen, especially when it comes to dealing with the representation of time discrete signals, as in sampled signals. Quick and dirty tests like this that shows such obvious errors are on my top list of sceptical subjects. If there is such an obvious error, don't you think the manufacturer of these plug ins would have noticed as well?
Loaded up the (freely downloadable) Span analyzer and played around with it, sine waves and the L2. Was at first unable to duplicate your results, until setting the window options exactly as you did.
Then I tried various frequencies. At 900Hz, there's almost nothing left of all this distortion. Looks rather nice at 1050Hz too. The display is horrible around 1000Hz!
Sound Forge's FFT showed much the same artifact. Tried null-testing and it failed at those frequencies. The remaining waveform was very low, at the level of 16 bit dither, and showed a nice sine wave cycling up and down in amplitude. Another hint of discrete sample point weirdness.
But the null test did not fail exact integers of the sampling rate. Ie, at 1102.5Hz or 551.25Hz, the original and L2 processed waves nulled completely!
This stuff only happens when the L2 (or L3) as active though, the display is as clean as can be when the threshold is too high to affect the signal. So there must be something happening, but I'm not sure if it's creating distortion or not.
This is way beyond my head, especially first thing in the morning! Think it's got something to do with the shortcoming of FFT. This link was interesting:
http://www.steema.com/FFTProp/FFTProperties/FFTProperties.ht m but I can't mesh that info with these tests right now, my brain is hung ower at the moment.
Someone with the knowledge of FFT's, sines, processing and sample points care to step up and explain?
Andreas
EDIT: Did another test on this weird limiter behaviour. Viewed the DAC outputs on an oscilloscope, one channel being the clean sinewave and the other with the L2 engaged as you did in the test, 3dB limiting. No difference.