PCB clock traces are not all that great most of the time, Intel or not, but it isn't that hard to get them more than good enough. Cost is a big driver (even for prototypes and reference designs). Accuracy is determined by the source, which is usually either a packaged oscillator or crystal circuit, and either way can be more than clean enough to meet spec. The usual issues with clock traces are noise and crosstalk coupling in the board (and devices) plus loss and discontinuities in the traces since clock distribution networks tend to get large and things like PCIe edge connectors are not well-matched from an RF standpoint. Adding an external source using RF cables may well provide a cleaner clock but I just don't really see the need. Plus most of the clocks are differential so you need twinax cables (or a pair of RF cables if you are loosely coupled). And it does not solve the problems of loss and crosstalk to other components on the board, just brings in a dedicated clock for the USB board (which seems like all you would need, but maybe you have to replace them all to maintain synchronization among clocks -- not my area of expertise).
As for the connectors, I would guess the hack involves finding a suitable place to hardwire a cable. I have seen some boards with test points to SMP connectors but usually those are not populated and only used during initial test and validation.
Still seems like a solution in search of a problem; is there a link to a description of what this hack brings to the table, i.e. what improvements it claims to provide?
