Yes, this works. In fact it's how a lot of testing of 802.11 gear gets done.
There's one critical thing you're missing from your described setup, and that's sufficient RF signal attenuation so your receivers aren't completely overloaded by your transmitters. You generally need about 60 dB of attenuation between transmitter and receiver. If you have the 3-leg (via splitter) scenario you described, you should put 30 dB of attenuation on each leg, so that all 3 paths end up with 60 dB of attenuation end to end.
I should elaborate on my "60 dB attenuation" rule of thumb.
That's based on the assumptions that most Wi-Fi gear nowadays uses +20 dBm transmitters, and that -40 dBm RSSI is a plenty-strong signal that should allow you to get maximum data rates without overloading your receiver. I've seen poorly-designed Wi-Fi cards that overload not much above -40 dBm RSSI, but I'm told that well-designed radios should be good up to -20 dBm RSSI. -10 dBm RSSI is probably pushing it.
If I were designing an RF cable rig for Wi-Fi testing, I'd shoot for max RSSI values below -20 dBm but above -70 dBm, maybe above -65 dBm. Somewhere around -70 dBm RSSI is where most clients start to be unable to maintain their maximum data rates. If you're designing something with a programmable attenuator or step attenuator to simulate different signal strength conditions, and want to be able to simulate being on the hairy edge of the network and losing signal, make sure you have a way to get to -95 dBm RSSI or less (so with, say, the maximum FCC-allowable +30 dBm (1 Watt) transmitter, you'd need 125 dB or more of path loss. You won't be able to get that much path loss just by putting attenuators in the RF cables. To achieve that much separation, you'll need to put the devices in separate shielded boxes, because the cables and the antenna connectors and things leak a fair amount of signal.
Update 2: Getting tired of trying to make this fit in the Comment box. :-)
@31eee384 In your comment where you mention your 1-to-4 and 1-to-8 splitters, you're close to getting the math right, but it sounds like you didn't take the insertion loss of the splitters into account. You may need a brief lesson in RF power splitters.
In, say, a 1-to-2 splitter, the "Sum" port (the "1" in "1-to-2") is different from the other two ports (let's call these other two ports "A" and "B"). The goal of most splitter designs is to get as much signal as possible from Sum to A and from Sum to B, and as little signal as possible between A and B.
"Isolation" is the attenuation between A and B.
"Insertion loss" is the attenuation between Sum and A, or between Sum and B. Because this is a passive splitter, power from the Sum port is split roughly evenly between A and B. Remember that in dB, adding 3 dB is a doubling of the power, and subtracting (or "attenuating by") 3 dB is a "halving" (cutting-in-half) of the power. So if you take a given input signal on the Sum port, the A port only sees half of that. We say that there is a 3 dB a "insertion loss" introduced by inserting the splitter into your RF path.
Splitters aren't perfect, so there will be a little more insertion loss than just the unavoidable loss inherent in splitting up a signal n ways. So the insertion loss of a specific splitter might be 3.2 dB, which is sometimes quoted as "0.2 above 3 dB", which calls out the fact that even a perfectly ideal 1-to-2 splitter will have 3 dB of insertion loss, and the 0.2 is in a way a measurement of how much (or little) they've varied from the ideal.
A 1-to-4 splitter splits up the signal 4 ways, so each port gets about a quarter, so an ideal one has 6 dB of insertion loss.
A 1-to-8 splitter splits up the signal 8 ways, so each port gets about an eighth, so an ideal one has 12 dB of insertion loss.
So in your example, the two clients on the same 1-to-8 splitter will see at least 30 + 22 + 30 = 82 dB of attenuation between them. Good.
Now what about the attenuation between a client on one 1-to-8 splitter and a client on another one? It would be 30 + 12 + 22 + 12 + 30, or 106 dB. You happened to get close with 104, but I think you got that answer the wrong way.
Okay, so what about attenuation between a client and the AP? It would be 30 + 12 + 6 + 30, or 78 dB. Even if you only had wimpy +15 dBm (32mW) transmitters in your AP and clients, that would still come out to an expected RSSI of -63 dBm, which should be good enough to maintain maximum data rates. And if you had full 1 Watt (+30dBm) transmitters, you'd be at -48 dBm RSSI, which is fine. It sounds like you'd even have room in this design to reduce or completely remove the attenuator between the AP and the first splitter.