>MSDN is hardly common, and isn't licensed for production anyways. Useful for devs but that's about it.

Dunno about small & middle sized outfits, but at the insurance company I optimized the code for a number of years they were allowed to build different "golden discs" to install a licensed # of machines with. They had licenses in parallell for W98, W2K and XP early this century for different use case scenarios. Dunno if those license deals evaporated after a certain time or were forever as MSDN back then, on the hope that the bulk of machines would always go to the newest OS deemed safe for the concern but a couple of dedicated machines could be re-installed the old OS if something broke on the HW side. I had the impression they had perpetual licenses, but back then there was no reason for me to verify... Flip side of that approach was the effort to test/verify on the HW platforms allowed in the concern, as machines not following one of a few family blueprints might miss some drivers and misbehave.

>
>The footprint issue is well taken. For quite a while I ran NT 3.51 on 32MB of RAM (cost me C$1500 in early 1996 just for that much RAM) on a P133.
>
>Then upgraded to an Athlon/1000 and 512MB RAM. Ran W2K on that initially, then Server 2003 R2 EE (same code base as XP) in Workstation (stripped-down) mode. The latter ran surprisingly well, usually with just under 256MB RAM in use with nothing running. So 384MB would have been painful and 256MB pretty much unusable.
>
>There were some issues running later-version VFP on NT4 i.e. http://fox.wikis.com/wc.dll?Wiki~VFPSupportedOS~VFP so W2K is probably a better minimum OS. W2K recommended RAM is 64MB (supposedly 32MB minimum) https://technet.microsoft.com/en-ca/library/cc977140.aspx which is a lot smaller than XP. Another bonus running as a VM is host RAM can effectively act as a disk cache for multiple VMs at once, making them run a lot faster.
>

Unless VFP9 Sp0 is used, I doubt any real problems can occur on NT4. Support for vfp today might be hard to get on any OS - chances for finding problems are much higher ;-)) W2K for me had on physical machines the HUGE benefit of stable USB drivers. In a VM not that noteworthy as you can map access to real USB devices via the newer host OS

>>>Another tool worth mentioning/consideration in some circumstances is running a VFP app in Compatibility mode. You might recall I ran into an obscure bug running VFP9 on Windows 10 some time ago; until I could implement a workaround the temporary fix was to have users run the app in Win7 compatibility mode.
>>
>>Good point to mention. Still: using (better: being forced to use!) such options feels to me like the aequivalent of monkeypatching code to "fix" bugs without going to the bottom, as you have no idea whether the "fix" will still work after the next OS patch day and in a perfect world you had time and budget to test after each patch day....
>
>Well, if you can't get the licenses and/or installation media (or support, for that matter ;)) then Win7 or Win10 it is. Tech support then either has to hope nothing breaks, or has to develop some sort of validation test/suite and run it against each new release :(

Yupp, that useless endeavour (unstoppable for real/long on Win10) might have prompted or fueled the whole train of thought ;-)
Set up once the old VM machines, make SURE they are isolated and stop scheduling tester time after each patch day