OK. Turns out neither is correct: EIP isn't truncated in PM to VM IRETD. But what neither emulators(Dosbox AND PCem) do is check EIP against the new CS limits and throw a #GP(0) fault if it does(on the IRETD, not on instruction fetch(PCem) or invalidly truncating and not faulting(Dosbox)).

Just implemented that in UniPCemu.
Also noticed that #GP(0) wasn't thrown for CS:EIP limit violations on CS loads(e.g. JMPF/CALLF/RETF/INT/IRET loading CS) in Real and V86 mode.
Perhaps that's what was causing software like CheckIt Diagnostics and WhatVGA to fail in real mode?
Edit: Nope. The bugs still occur(runtime error 200(WhatVGA) at IP=00AD and runtime error R6003 - integer divide by 0(CheckIt)) 🙁

Well, at least V86mon doesn't double fault anymore. Now i'm getting a Fault/Exception #13(0000)@0030:008B after which it halts the whole system.

Windows 98 First Edition still crashes during normal boot, but I see it crashing executing segment 1E? That's kind of weird for real mode, since it's in the middle of the IVT?

OK. Segment 1E is kind of strange in real mode, as that's in the middle of the IVT. So the issue is in the place it's called, which is at segment 70h. But isn't segment 70h MSDOS.SYS usually? Or in Windows 9x's case, definitely IO.SYS(since it doesn't have MSDOS.SYS as a boot file).

OK. The issue with V86mon was that the executable was corrupted in the disk image. That at least solves that issue 😁
So that means that the Pentium is at least running fine in V86 mode and VME's V86 mode.

So there's still some strange issue with real mode and various software(CheckIt, WhatVGA and PC-Check 6).