In this thread (Using HCSR 04 for communications) I posted some comments about my experiments in communications via ultrasound that you might find of interest. The issue I see with what you've proposed is that there will be echos from any transmit device that will greatly confuse the interpretation of the received signal at any site.
At least two approaches come to mind. First would be to perform signal processing on the received signal prior to detection to determine and remove the channel characteristics (the echos). This would require far more processing power than a typical microcontroller and non-trivial mathematics.
A second approach that might be workable is to have a known station-dependent delay on each of your receivers that allows the echos from any previous transmission to die out prior to responding. For example R1 might, upon hearing a pulse, disable listening for 100 milliseconds and respond after 10 milliseconds. R2 would do the same except it responds after 20 milliseconds and so forth. T1 then calculates the round trip time to R1 as the time the first pulse is received in the window 10 to 20 milliseconds after its transmission minus 10 milliseconds, and so forth . . .