While interesting, the conclusions just aren't true. Here's one more in favor of in-house.
Integrating the vendor solution will be more complex for the sake of unused feature over a simpler solution developed in-house.
Another: The solution to our problem doesn't use the other pieces we want/need to integrate, or doesn't exist. There's several solutions to manage docker clusters now, and effort has been made, but when they all started (some originally in-house) those solutions weren't widely available.
Another: the existing vendor solutions are sub-par for our needs (and/or the needs of our customers). Could you imagine if Joyent had just used VMWare as it's virtualization solution instead of the Jails from their OS based on Solaris? They have the only solution with really locked down Docker.
Sometimes creating a new trail really is the better solution. I wrote a test LMS a bit over a decade ago, that worked well enough for a lot of integrations before being replaced by a big vendor's LMS and it cost more for the integration alone than the customizations to the one I built... 10 years of service cost less than the 1 year of integration.
