Tested some more boards :
2x Gigabyte X58A OC : boot 135-136 trainable 138-139 setfsb both 140 , batch numbers (RT,ITC,FQC and lot ) are different compared to the "golden one"
Foxconn Flaming Blade GTI : boot 121, trainable 137, setfsb 137
Since nearly every board is somewhat trainable next to around 138 there should be some kind of synch tolerance every train step (it is 2 to 5 pcie steps on all the boards, except the golden one, there i was able to boot up 135 with can go straight to 150).
So maybe it is just the quality of the clock gen (can not find data for all but in other clockgen data sheets you will find the maximum range is not guranteed), the northbridge or just the ram detection / routine in combination with the board quality. Playing around with the ram settings (especially turn around settings) is next, but since the C1 Debug is not moving after bootup on gigbyte boards with pcie to high, i think it is a detection problem.
Something to look for was that maybe the Intel Mangement Engine is the kicker here, but digging around this it seems that it is not implemented in X58 consumer boards ,despited me_cleaner project is claiming it should be in some kind, but size of the bios files and the spi is also saying no, data sheet also says you can set a flag to run the chip without ME.
Later this day i will do a dump of the spi to check, but i think it is a dead end.
I use a Samsung 950 Pro with a PCI GT 610 for testing, looks like the Samsung Controller can take as much PCIe as i wanted him to take (i think since it is a PCIE 3.0 Controller and we running at 2.0 there is high tolerance due half bandwith cut).
Downside : currently only bootable for Windows 7 or higher, with XP the legacy option rom of the SSD conflicts while booting and i think there is more of a problem too (win-raid.com is the place to dig arround this). Conflicts with other onboard controllers (sata stuff) are common so you have to turn them of if you test it yourself. Also the adpater you are going to use is important (will post this later on).
Also tested an AHCI based PCIe SSD to boot XP(first gen of ocz revodrive) but this thing faded out at about 130 (it is a PCIE 2.0 device, sandforce controller with a raid controller too). There are some other AHCI PCIe SSDs X58 bootable around (Plextor, Kingston) but since they are all pcie 2.0 standard i guess you will run into the same high pcie problems here.
Using modern PCIE 3.0 NVME SSDs need an EFI Loader like DUET, that needs and extra firstboot device and i think is not an option.
So my conclusion at this point : Samsung PCIe SSD with PCI GPU is some easy way to go maxout your PCIE-Frequency (and looks reliable with above 150 PCIe) but not a must, since they are other ways to maxout your pcie clock to 140, above that it is going to maybe be relevant. Sadly the first board i tested with this setup was the "golden one", so i was a little bit too happy about and thought this is the way to go ?