My best guess is that there is a 90% chance Apple will wait for the Xeon v4 ...
What is suppose to be magical about Xeon E5 v4 ? There will be a bigger jump between v2-> v4 than v2 -> v3, but the delay is also likely twice as long. ( v3 could 'go' in December and v4 will probably be another year after that ). The supporting chipset is exactly the same ( Intel doesn't usually change workstation chipsets for an entire tick-tock cycle).
v4 clocks will probably go up and the top end, "if you have to ask you can't afford it" core count will go up, but there is likely not going to be radical change in TDP. v4 is 'tick' ( primarily process shrink) iteration. Up in the Xeon E5 zone there no mobile specific process tech or GPU updates that Intel is focusing on in the v4 ( "Broadwell") upgrades in the mainstream (e.g., Core M and broadwell GPU ). More likely Intel will use process improvements to crank up performance through more "turbo" and additional core counts. (**)
Same socket and chipset means the PCI-e lane count is exactly the same on v3 and v4. So still have same log-jam of moving to something like Thunderbolt 3 (even if it does come another year out ) or moving to multiple internal SSDs. Unless waiting on some Q4 2015 or 2016 components the same motherboard for a Xeon E5 v4 Mac Pro could also use a Xeon E5 v3. So waiting on what new infrastructure?
A 90% weighting on "Apple will skip Haswell (v3) like they skipped Sandy Bridge (v1) " doesn't particularly have that kind of firm grounding.
Kneecapping the Mac Pro with Xeon E5 1620 v2 CPUs while the high end iMac moves to Broadwell on the high end is only going to cause Apple incrementally more problems with Mac Pro sales. Even more substantive impact for the whole Mac Pro line up when the other workstation competitors move on.
A huge delay into late 2015 for the Mac Pro would be far more indicative that Apple has nobody working on the Mac Pro full time more so than the Intel tech coming in v4 was a show stopper issue.
(**) Intel has a growing roster of low TDP server parts. Xeon E3 . The new Atom Avoton and 14nm shrinks of Avoton. (
http://www.theregister.co.uk/2013/07/23/intel_lower_power_xeon_atom_server_chips/) There little to no reason to radically change the Xeon E5's TDP.