Ah, no it doesn't really work like that, an RF attenuator is usually a matched impedance device, that is to say it is a Pi network 50 ohms in and out.
If you put the attenuator in parallel with a dummy load you end up with each taking half the power and the rig seeing a 2:1 low Z VSWR at least until the attenuator melts.
What you really want is a dummy load with a sample port (This is really a sort of attenuator, but one designed to handle big power on one port), they are usually 30dB or so. You follow this with one or both of your 20dB attenuators, no reason to be anywhere near the upper limit of the SAs power handling.
A warning, high power attenuators (Same thing as a dummy load with a sample port) are often not really bidirectional, they tend to have the high power resistor only on one of the two ports, this is a trap for young players as they often have a N type on both ends (Who? Me, 30 odd years ago? Never)....
You can build a sampling tap that would go between your rig and your existing dummy load, use a T piece on the front of the dummy load and follow it with a network consisting of a 2.5k resistor in series and a 51 ohm resistor as a shunt across the sample output, this will be good for pretty close to 40dB at least at HF, things get harder in UHF and microwave. Build this network with short leads on a bit of copper clad.
+46dBm is 40W, most radios are designed for 100W PEP output, so I would assume 100W (the extra will not hurt as you will be using the SAs adjustable input pad anyway and usually want about -10dBm at the SAs first mixer).
73 Dan.