I think there are only 3 key questions regarding data transmission protocols: bit perfect transmission, jitter and noise.
Bit perfect transmission is a complete non-issue, except some protocols may support higher sampling rates than others, but they are all adequate for most folks. There is also the question for some, like me, of support for Mch formats without multiple runs of the interconnection protocol. Except for those issues, which are possibly significant for some, none has a general advantage here.
With jitter, Amir has supplied pretty good measured evidence that at least some, though not all, implementations of each protocol would seem to offer audibly insignificant amounts, though possibly not with HDMI transmission, due to its control by the video clock. However, I do not think he has yet measured anything via I2S. Has anyone? But, still, it may be technically harder and/or more costly for a maker to reduce or eliminate jitter with some protocols than with others. Offhand, though, I don't recall seeing any measurements of asynch USB that seemed to suffer excessive audible levels of jitter, but possibly some implementations of it screw that up. And, with I2S, there is always the question of jitter induced by the long, external clock leads.
Noise transmission that interferes with the D-A to me seems like just a DAC issue, not a transmission protocol issue, although some insist that it must be painstakingly eliminated upstream and/or via gizmos in the transmission path. But, for some reason, there still seem to be plenty of DACs that have schiity noise rejection. Does Toslink have a significant advantage here? I doubt it vs. a properly isolated DAC, and Amir has numerous measurements of DACs metallically connected that show no significant problems. I also just love DAC makers who make and recommend their own addon de-noiser along with their DAC. It's like a reviewer guy recommending that you use an after-market, audiophile power cord with some electronics, duh. Did they forget someting in their original design?
So, what else could we possibly need? I will add a 4th key issue, which is wide availability of the connections for the protocol at both ends to the commonly used devices. Add to that relative ease and lowest cost for effective implementation. I think here is where we see some separation between different protocols. I will not score all common protocols on that, because the answers are fairly obvious. But, it should be noted that the non-standardized I2S protocol has a problem here.
So, in light of the thread topic, here I am with the best sound of my lifetime using asynch USB in hirez 7.1 Mch up to 384kPCM or DSD256 with up to 7.1 channels into a galvanically isolated DAC, but simply sourced from a noisy stock PC via simple plug in to a connector that was already there on the PC, actually multiples of it, and all with proper plug'n play drivers. Why on earth would I want to consider something else, like I2S requiring added hardware and driver software on the sending end in the PC?
Is there some other special sonic magic, better measurements, etc. about I2S that overcomes its practical limits in cable length, interconnect and plug standards, sharply reduced DAC selections, etc. What is the "pot of gold" for going to the trouble with I2S vs. plain old, highly performant asynch USB into a wide choice of different, well designed DACs at a wide range of price points?
Sorry, but, I am still seeing little but a mythology, an "anything but USB" dogma based on precious little actual evidence that I2S is in any way an improvement beyond asynch USB.