Also, mismatching down is harder on a tube amp than mismatching up, within reason of course. If you have an amp that wants an 8 ohm load, and you put it into a 4 ohm load, it's halfway to a short, and you run a higher risk of the output tranformer arcing. If you go the other way and hook up a 16 ohm load to an 8 ohm tap, the tubes are actually running cooler and will last longer. Go either way from the optimum and the volume is less and the distortion is more. You wouldn't want to have a mismatch more than one step either way or you could have some failures.
I do agree that proper matching of load and output impedance is the best for tube amps - it's the only way to get the maximum power and minimal distortion out of them.
However, the first part concerning a proper direction of mismatching is backwards.
Many people believe a "short" is a bad thing and should be avoided at all costs - after all it should be "common sense"... right?
Unfortunately this is simply not true for a tube amp with an output transformer - where running it with no load, meaning a completely open secondary is the fastest way to make that output transformer into something suitable for a boat anchor.
But I am not going to prove it with theory that no one will read - someone else can chime in with that, it's boring anyway. I'm going to give you a little history instead.
If you ask almost any amplifier tech which tube amps have the most and least frequency of output transformer failure, 99 times out of 10 the answer to the most will be Marshall and the answer to the least will be Fender. What could possibly cause this difference?
Fender built cheap amplifiers. No, this is not a statement that cheap is better, it's just a statement of fact. Leo was a businessman that knew building an amp cheaper would make his company more of a profit.
For example: traditional Fender output transformers only have one secondary and not the multiple taps you will find on Marshall amps. One of the primary causes of the difference in failure rates was the rather cheap impedance selector plug used on early Marshalls - if you didn't get it in just right it could fall out, opening up the secondary load on the transformer and making it history.
During the 1950's Fender made combo amps. Marshall chose the 1959 bassman schematic for the basis of their amps and incorporated various improvements like the tapped output transformer mentioned above.
During the early 1960's... Fender started making piggy-back versions of some of their combo amps like the Tremolux, Bandmaster, Bassman and the Showmans.
Unlike Marshall, Fender recognized a problem with this arrangement: unlike a combo amp where the risk is rather low, with a piggy-back configuration the chances of having someone power up the separate amplifier head without a speaker load attached was much higher - and having to replace output transformers under warranty would significantly eat into the profits of the company.
To minimize this risk, Fender made a small change. A change so small it went un-noticed by most of the amp making world that copied their designs - especially Marshall. Fender spent a few extra cents on the speaker output jack and wired it so when no speaker plug was in the jack, the secondary of the output transformer would be SHORTED instead of OPEN. This modification was eventually incorporated into the entire tube amplifier line, combo amps included.
You can check this for yourself by browsing the wealth of Fender schematics available at the Fender Field Guide (or most any other place on the web for that matter).
I mentioned that the Fender output transformer isn't tapped with various impedances. It's secondary winding impedance is for the "normal" load. For example, the output impedance of the secondary for a Twin Reverb is 4 ohms and it's two 8 ohm internal speakers are wired in parallel for a total of 4 ohms.
The Fender extension speaker jack is simply wired in parallel across the main jack. So anything connected to this jack would be in parallel with the main load - increasing the load by reducing the load impedance. A 4 ohm extension speaker cabinet on a Twin would give 2 ohms total.
If this increased load... this "closer to a short" was indeed the worst case scenario for an output transformer then Fender could have easily wired the speaker jacks to put the extension speaker load in series with the main load.
But they didn't. And it is well known in the world of Fender amp users that the amps can easily deal with a 100 percent mismatch downwards.
I give you the wild man of rock himself... Mr Ted Nugent. For a significant part of his performance career he used multiple Twin Reverbs each on top of and with it's extension speaker jack plugged into... a Dual Showman cabinet - a 2 ohm load with a 4 ohm output impedance amp.
I don't believe Ted knows any amp setting other than "10". I've seen him perform something like 5 or 6 times with that setup - and never once did it fail. If it was an arrangement prone to failure, I'm quite confident he would have changed it for something more dependable.
It's your amp and you can run it any way you want. If you believe a mismatch upward is best and the output tubes will run cooler and last longer, then by all means go for it.
And if you do blow the output transformer, which is much more likely in this direction of mismatch, just remember how long your dead cold tubes are lasting.
