Just for interests sake, a common 1/2W, 1% tol., 2.5mm diameter metal film resistor has 1/3 pf parallel capacitance in free air according to my measurements. I went about measuring it because an amplifier prototype for my analog computer using a 1M ohm feedback resistance was suffering undue reduction in phase margin due to the 0.33pF of capacitance, which corners with 1M at about 480kHz. I got around the issue by making 1M with five 200k in series instead, lowering the net parallel capacitance to one fifth of one third of a puff.
BTW, I'm not sure what you mean by "higher thermal noise of the 25k resistor". If the characteristic impedance of the attenuator is 50 ohms, the noise generated will be equivalent to 50 ohms, regardless of how the attenuator is made.