The reason why PCM1704 audio-gd DACs are limited to 96K is because of the oversampling filter.
The PCM1704 can accept up to 768KHz. Given that Audio-gd is using an 8x oversampling filter by default, it means that the maximum acceptable rate is 8 x 96 = 768 KHz.
The DSP1 digital filter can accept up to 192KHz. In that case, the upsampling should be set for no more than 4x (4 x 192 = 768). Doing that would enable playing back 192KHz files but would reduce the playback quality of all the other files (44.1, 48, 88.2, 96). Most people who have compared the 4x and 8x oversampling settings in the DSP1 preferred the latter.
There is also the issue of the Digital filter. Kingwa has compared both the 24/96 DIR9001 and other 24/192 digital fitlers (such as the very commonly used CS8416) and found that the DIR9001 was superior souding.
From all of that, it seems that in order to accomodate for 24/192 playback, it would involve one or more compromises (reduced oversampling and/or lesser digital filters) for something that might not even be beneficial.
I remember reading somewher that Dan Lavry said it is pointless to go beyond 24/96 and that many DACs behave poorly at 24/192. Benchmark (and a few others) are setting the upsampling rate to 110Khz instead of 192KHz because the DAC they tried actually perform better at those sampling rate.
That is not to say that 24/192 is bad. But it doesn't necessarily mean that a move from 24/96 to 24/192 will always yield an improvement.
Some say that anything above 16/44 is pointless because human ears can't hear beyond 20K which is below the 22K frequency that CD can reach. While it is true that we don't hear above 20K in the frequency domain, using a CD or DAC at 16/44 causes time domain and phase distortions that fall back in the audible range. Transients amongs others get smeared at 16/44 especially with "regular" linear phase/fast roll off filters. The DACs that don't have a gentle roll off before 20K are actually the worst sounding and the most poorly designed. That is why we have seen the rise of new filters that use minimum phase and apodizing filters that sound more natural to our ears.
Those who still believe that anything beyond 16/44 is useless, I invite you to read Kunchur papers on the subject of temporal resolution (here:
http://www.physics.sc.edu/~kunchur/Acoustics-papers.htm). In his testing of the limits of human temporal resolution, he couldn't find a CD Player that was good enough (or accurate enough) for the testing. He high to use a high resolution analog wave generator for the testing. If the CDs (16/44) were "perfect forever", why wouldn't be sufficient for the testing?
To sum up, I would say that the reason is valid regarding the limitation at 96KHz. I believe that most libraries out there are composed of 44.1, 88.2 and 96, and it would be a bad idea to compromise the playback of 99% of the files (for the majority) in order to make it possible to "read" 24/192.
As an example, I would rather listen to MP3s on my dac19dsp than to 24/192 files on my emu0404usb.
The new NFB series that use the ES9018 sabre chips are a little bit different. The ES9018 can accept 24/192 and are much less prone to jitter (in paper) than most DACs out there. I wonder why Kingwa hasn't given the possibility to use one the new 24/192 WM8805 digital receivers instead of the DIR9001. Even if there is a trade-off in jitter quality (which might not even be the case), it wouldn't involve a trade off in oversampling/filtering like it is the case for the dsp1/pcm1704uk combo...
