I'm out of town (in Virginia) for 10 days and don't have all my resources handy. But if you find the Cornell-Dubilier web site they have some extensive information on various capacitors, including the effect of heat on lifespan. Bottom line, there is no such thing as a magic temperature marking the boundary between exploding and living forever... cooler is always better.
For what it's worth the 2A3 direct coupled version dissipates much more heat from the cathode resistors, because the cathode is at 260 volts.
Capacitor value: my estimate for the best value is 2*L/R squared - L being inductance and R the transformer primary impedance. At this value, the impedance presented to the tube drops slightly (2/3 the midrange value? Sorry I don't have my simulations at hand) in the deep bass, then rises as the capacitor dominates at lower frequencies. Note that some Japanese audiophiles use much larger capacitances. Even the largest cap will perform much like a series feed arrangement; one of the parafeed virtues is a more resistive, higher impedance load presented to the tube at the lowest frequencies.
A smaller capacitance raises the cutoff frequency (roughly as the square root of the capacitance), but provides the tube with a safer minimum load. Somewhere around half the nominal capacitance the load never drops below the midrange value. In general I think this is a good compromise for using the 45 because it keeps the distortion low at maximum power. Smaller than that, you lose bass extension without getting any benefit in return.
Conversely, a larger capacitance will deepen the bass extension for small signals, at the price of possible distortion with large low-frequency signals. Since I run 101dB speakers with 300Bs, I use a 6uF cap in my Paramounts - I'm not likely to run out of steam. Around twice the nominal value, the load impedance benefits become small relative to a regular series feed.
Keep in mind that this assumes a perfect power supply, and a speaker impedance that is a pure 8 ohm resistor. There will be some interaction with both impedances, so the generalizations above should be taken with a grain of salt. I recommend experimentation with the target speakers in the target room - listen for bass distortion and extension. Once you know what capacitance works best, then you can spend some bucks on higher quality caps to get cleaner treble. I'd get a ~6uF cap, and parallel the 3.3 to get 10uF - gives you three values to listen to.
Well, not the answer you were hoping for, but food for thought. So much of engineering is not answers so much as the art of compromise...
