For the record, my 6,1 (D300, latest firmware) won't boot with a single eGPU (GTX 1070 Ti [Pascal] in Sonnet eGFX via Apple TB3-TB2 adapter) connected to any Thunderbolt bus other than the first, even with both factory GPUs in place, but has no problem booting when the eGPU is connected to the first Thunderbolt bus (bottom two ports).
Moreover, my 2012 Mac mini server (i7, no factory dGPU), with the same eGPU attached, boots if and only if there's also a display connected to the built-in HDMI port, but has no problems booting with neither display nor eGPU connected (!?!).
So without any special knowledge of the firmware itself, I wouldn't personally make any assumptions about how Apple firmware reacts to unsupported GPU configurations without actually testing the particular configuration against a particular firmware version on a particular model.
Come to think of it, I'm not sure I'd even limit this to unsupported
GPU configurations: with no eGPUs in sight, this same 6,1 only rarely (and seemingly randomly) displays boot screens on the Samsung 3440x1440 connected (via DisplayPort) to the
factory GPUs. Moreover, in the rare event that boot screens
do display, they do so at full native resolution, so that, at least, is not the problem.
And the Mac mini — again, no eGPUs involved —
always shows a boot screen on this same display, albeit at a lower-than-native resolution (even though the iGPU itself also supports the display's full native resolution).
Given these results, I suppose one could just as easily speculate that the 6,1 firmware may also whitelist
displays…when reality is probably far less interesting (unanticipated configuration exposing a latent firmware bug, say, that Apple isn't particularly inclined to address because "unsupported").