They made tubes up to 1.5 MW, watts are not a problem. Now, if you expect it to be compatible with a 120V 15A circuit, or 240V 7A circuit, don't expect more than about 800W real output (even for a solid state linear amp), more likely 400 or thereabouts (or about double that from a 240V 13 or 15A circuit, etc.). Probably the Fender example is such a case.
What tubes to use? Doesn't much matter. 6L6s and the like (more or less including the families up to KT88s and such) are preferred for audio to begin with, and are cheap (relatively speaking) and plentiful in current production. 6080/6AS7 is hard to drive, more expensive(?), has lower plate dissipation, shorter load line (triodes have less power output than tetrodes), and consumes far more heater power. Sweep tubes are generally expensive, and offer high plate efficiency but only modest overall efficiency (again, heater power).
Sweeps are generally quite powerful, especially when driven hard, as befits their original application. One could construct a class D tube amp, doing probably over 800W (from a 1800 VA circuit), with just a quad of, say, 6LF6 or 6LV6 or EL519 (or any of their series-heater brethren). The power density easily offsets the cost of obtaining these tubes. One might (rightfully?) question whether such an amp would be worthwhile, i.e. create "That Tube Sound", in the presence of so much distortion (PWM modulation is essentially "100%" distorting a signal plus carrier). Note, you still need an output transformer rated for the full signal bandwidth -- you don't get to save any transformer size with just PWM alone, not without a lot of extra work (see Dave Berning's DC transformer / chop-amp designs for example).
Tube cost isn't a huge motivation, anyway; it is a maintenance cost, but the rest of the amp will easily cost upwards of $1000 to build or buy, too. The capital/operating cost tradeoff is a worthwhile concern here. One could go with NOS industrial or PQ types, derated modestly, to obtain a likely 10khr+ lifetime, versus cheap Chinese 6L6s that will wheeze out in a tenth of that time. Derating requires considering this in the design -- you might only need 8 x 6L6s if you're beating the shit out of them, but 10 or 12 might be needed to give adequate derating.
Protection circuitry is a good idea regardless, as that beefy output transformer won't be something you want to melt from an errant sparking tube. Adequate fusing and/or supply limiting, and maybe bias servo control, are simple enough precautions.
The power transformer doesn't need to be expensive, fortunately. A switching supply is very reasonable at this power level; of course, making one that operates safely and reliably without belching out EMI, may be best left to the professionals...
Tim