We ran the same server for nearly 10 years until the original drive failed.
It had had 2 or 3 extra drives, 3 NOSes, 1 case and 3 CPUs and mother boards.
Last seen with one of the newer drives hosting a 4th NOS.
Run out of space - put in extra drive and merge it in, go a bit slow new MB processor - started as 3.11 ended up as 6.0 without original drive via 4.10 (I think) and 4.11 SP4
Except Netware was pretty rubbish at multiprocessor stuff until v5. Actually, didn't really get to grips with it until v6's Linux kernel.
Went well enough as a single processor - ours started as a 386!
Run 486 for a few years then a pentium of some form.
Back in the NW4.11 era multiprocessor was no big deal the server would handle more throughput than an NT box anyway - seen the tests ran by a supplier. Near enough twice the disk performance - about 4x total throughput for same hardware.
Tests were published - Extended Systems did them
I suspect a biased test. Remember I used to look after around 300 Netware (4.11 mostly) servers, and maybe 200 odd Windows (NT3.51/NT4) servers. Depending on users/application, mostly Compaq Prosignia 300 (small stuff) and Compaq Proliant 4500 (larger stuff) type stuff.
To be honest, not much between NW411 and NT3.5/4 performance wise. Netware lost a bucket load of performance between V3 and V4, yet the only real change was moving from Bindery to NDS. If you needed to use TCP/IP across your WAN routers, Netware's performance died even further, as your choices were IP Tunnels or the utterly awful NetWareIP.
Don't get me wrong, Netware was a decent NOS, esp with NDS. But Windows servers made as good file/print servers, and far better application servers. And more reliable to.
A supplier who produce NLMs and services found on a dual boot server that NT was about 1/4 the speed of Netware, they found the difference was mainly down to Netware disk handling being quicker.
I cannot find the test since they were bought by Sybase
NLM supplier - ie, trying to flog Netware stuff 
Either that, or set Windows up using FAT. NTFS has always been more efficient than Netware's filesystem, though its irrelevent, as all servers should be offloading onto a dedicated controller anyway. Back in the KnitWare 4.11 days, we were using either 16Mb (up to 12M battery backed) or 64Mb (56Mb battery backed) cached SCSI controllers, which were pretty much the norm everywhere. This jumped up to 256 then 512, then 1G caches. Generally now, for x86/x64 architecture, like most companies, we use VMs running on blades, with SAN as storage, so all caching done at SAN end.
Dual boot is the clue, as was the fact I mentioned services.
It was part of their FAQ to answer performance issues for cross graders.
They now do Linux as well but they did a database server before Microsoft did a NOS, so of course they started on Netware.
Our customers who have cross graded from recent Netware to new Windows have seen a performance drop but were expecting it. In all circumstances it came down to one thing - support.