When were the best tube amps made?

PrimaLuna comes to mind as a company that has combined the best of modern technology with the best of vacuum tube sound quality. Their autobias has taken the mystery out of maintenance and brought tube rolling to the market for those who would otherwise not want to deal with such things. At real world pricing their products punch way above their price point. Driving tubes gently helps keep the cost of long term ownership reasonable. Lots of reasons they’ve done so well in a short time. Perhaps others will take a queue. 

Sometime in the future.

I second PrimaLuna.  My EVO300 integrated tube amp is wonderful when upgraded with Golden Lion KT66 tubes  

With a single ended tube amp it is easy to build your own. But to keep it linear you can't push a tube very hard. You need heavy duty transmitter tubes. There is a WAVAC SET that sold for $350,000 that you can build with under $2000 in retail parts. It runs 833A triodes at about 1000 Volts which will make these tubes last for many years. Hammond makes a transformer more than robust enough and more robust than any bigger name or vintage transformers and Lindahl makes an input transformer for the grid suitable for a 45 SET to drive the 833A. 

This gets you SET purity you can use for less sensitive speakers such as planar magnetic speakers which have a purity that is difficult to approach with cabinet speakers. 

Transformers are the key element for both SET and push pull.  Even for OTL the power transformer is relevant.  Likely the best commercial winders were in Japan in the late '90s, companies like HIrata Tango and Tamura.  I think back then there was a focus on sound and quality rather than volume and profit margin. 

Likely there are currently some good boutique winders, and maybe Western Electric were good (never heard an amp with these).  Useful to distinguish between collectable and listenable - not sure where the WE falls.

I certainly do not equate 'high tech' with 'great sound'.  Frequently the application of high tech results in additional complexity and worst sound.  Hard to make the argument that the more you process the signal the better and more realistic it sounds.....