I guess I haven't addressed the impedance question very well yet - I was feeling pretty crabby last night (for which I apologize to anyone who might have felt offended. Completely unrelated reasons.)
Anyhow, in my opinion, or according to my analyses depending on how you like to phrase it, 4000 ohms is as close to ideal as you can get for a 45. A 3K load exacerbates the already poor damping factor of the 45. On the other hand, getting full power from a 5K load would require a higher plate voltage than the tube is rated for.
By designing at 4K, a range of about 2800 to 5600 will work quite well (I calculate that as the square root of 2, multiplied by or divided into 4K) - that is, they will all produce similar power and distortion, with a slight tradeoff of higher distortion and greater power at the lower impedance, and vice-versa at the higher impedance. In my opinion, this is the best available compromise; those who wish to use different iron are welcome do so as long as they can figure out how to mount the components safely and effectively - and don't expect Bottlehead to guarantee the result! :^) In many cases that will require a new, larger chassis plate - but that's always been the case, and probably always will.
In the original SR45, I jiggered the operating point (lower voltage and a little too much current) to work best with a 3K load, since that was the only available quality iron for the Paramour at that time. Stretching that to 5K was a bit much, so I offered some adjustments, mostly just reducing the current. It increases the distortion a bit, leaves the power unchanged, and extends the tube's life a bit. I'd like to stress that all the above are quite small effects - I think most audiophiles might hear them but would be hard pressed to choose which is best.
Incidentally, in my current draft layout there is more room than usual for capacitors. But as I said, I haven't even shown this draft to Doc B yet.
