If you have the ability to do an arc flash study on this panel, that would be preferable to trying to find a loop hole to avoid it.

Well technically a three-phase transformer is essentially 3 single-phase transformers connected together. They often share a common core.

But - I would still do the arc flash study on the panel and not be looking for a loop hole to avoid it.

The guideline is written for 3 phase systems. 3 single phase transformers banked together is equivalent to 1 three phase transformer in terms of the short circuit current available. I believe the intent of the one transformer statement is to exclude multiple sources.

These exceptions are all about the circuit and exception #2 "the circuit is supplied by one transformer" seems to mean anything over single phase; if not how can there be more than one transformer on a single ciruit?

The general reason for avoiding multiple transformers on a single circuit is that if the impedances are not fairly closely matched, you will get circulating currents, and because even though you may have two transformers in parallel, if the impedances are not very, very similar, electricity flows proportional to their conductances and thus one will get much more load than the other. This is not a simple case of the capacity of two transfomers rated X and Y = X+Y.

However there are certainly cases where transformers are operated in parallel for instance for redundancy reasons such as having 3 transformers in parallel with appropriate switching to allow any given transformer to be de-energized either due to a fault or for maintenance purposes. Granted I have never seen this with low voltage (<240 V) circuits but it would not surprise me if I did see it.

Second, I have pretty much come full circle on this issue. At this point the problem I have with IEEE 1584 in its current state is that it relies on a single data point at 208 V and that the documentation stated that they could not achieve a stable arc with any of the other tests. Arcing below 250 V is really unstable and I have seen research papers showing results that are both above and below the incident energy predicted by IEEE 1584. Further as Jim Philips who is Secretary of the committee has already stated publicly, the "<125 kVA, <240 V rule" which is really little more than a passing comment has already been shown to be inadequate and that there are definitely conditions where it does not suffice. So at this point considering the original poster's condition (granted this is months later) I would suggest using the values in NESC which are based on EPRI testing for the <250 V case. NESC gives a value of 4 cal/cm^2.

In the future I am at this point of the opinion that just as with the >15 kV case, we really can't use IEEE 1584 properly for low voltage cases. OSHA reached the same conclusion recently in their findings for the updated OSHA 1910.269 Subchapter V rules for utilities for over 15 kV. This probably means that we will be using a simple "table lookup" procedure for dealing with the low voltage cases.