No precedent was established as to whether their prior use actually was a derivative work - personally I've always shared the VMware take on it that this was a loophole, and that if the VMkernel was a derivative work of Linux by virtue of the vmklinux compatibility layer, then Windows becomes a derivative work of any open source driver, which doesn't make any sense...
Of course, when ESXi shipped without a console OS, but with a vmklinux and GPL'd drivers, a new take was needed.
It wasn't "close enough" - they released the drivers/patches/modifications of those drivers, per the terms of the GPL.
IIRC they also avoided at least one network driver (I believe Alan Cox's 3com driver?) because the author had strong personal objections, and 3com actually wrote an alternative.
They also open-sourced the vmklinux API/wrapper layer itself, so the full picture pre-ESXi was:
vmklinux module - GPL'ed module for a proprietary VMkernel
vmklinux-based drivers - GPL'ed forks of drivers from the Linux source tree
console OS - GPL/open-sourced light fork of RHEL
VMkernel - proprietary, closed source (later available under shared source w/ NDA)
Hellwig's lawsuit AIUI argued that there was significant implementation, covered by copyright, in vmklinux headers and elsewhere, such that the vmklinux<->VMkernel boundary was not sufficient to prevent the latter from being a derived work of the former.
I also think that beyond that there was a strong belief and assumption in the Linux community that other parts of the VMkernel "must have" been copied from OSS/Linux as there was simply no way a random/small/private company could have possibly created its own file system, scheduler, network and disk stack, etc.
But the reality was simply that VMware had a ton of really talented and highly productive operating system experts in the early days.
[I was an employee at VMware from 2002-2011, opinions my own]